ENRECAP roject Enhanced research capacity to meet VIETNAM future oil and gas challenges in Vietnam PETRO Petro ietnam
An Official Publication of the Vietnam National Oil and Gas Group Vol 10 - 2017
ISSN-0866-854X
EXPANDING OIL AND GAS EXPLORATION AND PRODUCTION AREAS PETROVIETNAM JOURNAL IS PUBLISHED MONTHLY BY VIETNAM NATIONAL OIL AND GAS GROUP
ENRECA Project Enhanced research capacity to meet VIETNAM future oil and gas challenges in Vietnam PETRO Petro ietnam
An Official Publication of the Vietnam National Oil and Gas Group Vol 10 - 2017
IISSN-0866-854XSSN-0866-854X
EXPANDING OIL AND GAS EXPLORATION AND PRODUCTION AREAS
EDITOR-IN-CHIEF Dr. Nguyen Quoc Thap
DEPUTY EDITOR-IN-CHIEF Dr. Le Manh Hung Dr. Phan Ngoc Trung
EDITORIAL BOARD MEMBERS Dr. Hoang Ngoc Dang Dr. Nguyen Minh Dao BSc. Vu Khanh Dong Dr. Nguyen Anh Duc MSc. Tran Hung Hien MSc. Vu Van Nghiem MSc. Le Ngoc Son Eng. Le Hong Thai MSc. Nguyen Van Tuan Dr. Phan Tien Vien Dr. Tran Quoc Viet Dr. Nguyen Tien Vinh Dr. Nguyen Hoang Yen
SECRETARY MSc. Le Van Khoa M.A. Nguyen Thi Viet Ha
DESIGNED BY Le Hong Van
MANAGEMENT Vietnam Petroleum Institute
CONTACT ADDRESS Floor M2, VPI Tower, Trung Kinh street, Yen Hoa ward, Cau Giay district, Ha Noi Tel: (+84-24) 37727108 * Fax: (+84-24) 37727107 * Email: [email protected] Mobile: 0982288671 Cover photo: Cam Lam - Nha Trang. Photo: Le Khoa
Publishing Licence No. 100/GP-BTTTT dated 15 April 2013 issued by Ministry of Information and Communications FOCUS FOCUS
Prime Minister Nguyen Xuan Phuc: Deputy Prime Minister Trinh Dinh Dung: PETROVIETNAM TO STAND FIRM IN THE DUNG QUAT REFINERY IS THE LEVER MIDST OF DIFFICULTIES OF QUANG NGAI’S ECONOMY On 19 October 2017, eporting to the Deputy capacity (106 - 108% on average). Prime Minister, General BSR has focused on implementing On 12 October 2017, at the Government’s headquarters, Prime Minister Nguyen Deputy Prime Minister Trinh RDirector of BSR Tran solutions to optimise production Xuan Phuc worked with key leaders of the Vietnam Oil and Gas Group to evaluate Dinh Dung paid a working Ngoc Nguyen said Dung Quat operation; promoting scientific the results of implementation of the production and business plan in 2017 and the visit to Binh Son Refining and Refinery has always been operating research, with primary emphasis deployment of key oil and gas projects. Petrochemical Co. Ltd. (BSR). stably, reaching the optimal on energy savings, and optimising
4 PETROVIETNAM - JOURNAL VOL 10/2017 8 PETROVIETNAM - JOURNAL VOL 10/2017 4 8
SCIENTIFIC RESEARCH
PETROLEUM EXPLORATION & PRODUCTION PETROLEUM PROCESSING PETROLEUM ECONOMICS & MANAGEMENT
14. Application of deep learning in 51. Testing antibacterial effect of 59. Energy efficiency in predicting fracture porosity silver nanoparticles Vietnamese enterprises: The on sulfate-reducing bacteria predominance of gas consumers 23. Different forms of Gassmann equation and implications for the estimation of rock properties 30. Merging of 3D seismic data 40. Quick pre-stack seismic inversion to predict reservoir properties at a gas and condensate field in Nam Con Son basin 45. Application of geochemical technique to reduce allocation cost for commingled production wells from multiple reservoirs CONTENTS
FOCUS PETROLEUM EXPLORATION & PRODUCTION
APPLICATION OF DEEP LEARNING IN PREDICTING FRACTURE Prime Minister Nguyen Xuan Phuc: POROSITY Pham Huy Giao, Kushan Sandunil Geo-exploration & Petroleum Geoengineering Program, Asian Institute of Technology (AIT) Petrovietnam to stand firm in the midst of difficulties ...... 4 Email: [email protected]
Summary
Deep learning (DL) neural network analysis is the latest development from the Artificial Neural Network (ANN) and it is being used more and more in petroleum engineering. In this study, the way to develop a new DL model for well log analysis was attempted and Deputy Prime Minister Trinh Dinh Dung: successfully implemented using well log data from a location in the Cuu Long basin, Vietnam. Three sets of analyses were conducted, i.e., the first analysis set with a single hidden layer ANN model, the second analysis set with multiple hidden layer ANN modeland the third with a DL neural network model. The DL-predicted porosity for a fractured granite basement reservoir of an oil field in the Cuu Long Dung Quat Refinery is the lever of Quang Ngai’s economy ...... 8 basin was found in the range from 0.0 to 0.082, showing a good match with the conventionally-calculated values. The final deep earningl model consists of 5-input layers of gamma ray (GR), deep resistivity (LLD), sonic (DT), density (RHOB) and neutron porosity (NPHI), having 5 hidden neuron layers with 14 neurons per layer. It is worth noting that the transfer function of the rectified linear unit (ReLU), typical for a deep learning analysis, was implemented to replace the common sigmoidal transfer function, ensuring the successful application of DL model. Last but not least, the problem of vanishing gradient specific for a DL neural network model was also explained indetails in PVEP and FPT co-operate this paper. Key words: Deep Learning (DL), Artificial Neural Network (ANN), fracture porosity, well log analysis, fractured granite basement, Cuu Longbasin. in research and development of new technologies ...... 11
1. Introduction complex version of artificial neural networks. It is now widely used in image recognition, voice processing and language translation. The concept of soft computing was first put forward by a paper named “Possibility Artificial neural networks are used to build a connection between theory and soft data analysis” [1]. While input and output data that helps predict the outputs of a set of new NEWS hard computing needs a precisely defined inputs (Figure 1). analytical model, soft computing is tolerant Application of deep learning in petrophysics is still in its infancy of uncertainty, imprecision, approximation stage. Classic neural networks commonly use one hidden layer, whereas and partial truth. This is a method designed deep learning uses multiple hidden layers. The use of multiple hidden to model and solve real world problems, layers can cause a phenomenon called vanishing gradient problem. Negotiations on Ca Voi Xanh project to be sped up ...... 68 which cannot be modelled mathematically. Deep learning network is obtained by eliminating this problem. The concept of soft computing was designed based on the concept of human Input layer Hidden layer brain functioning. X1 Σ Soft computing consists of few principal Nhon Trach 2 Power Plant components as listed below [2]: X2 Σ - Fuzzy logic; Output layer reaches production milestone of 30 billion kWh ...... 68 X Σ Σ k - Evolutionary computation; 3 - Neural computing;
X4 Σ Activation Transfer - Probabilistic reasoning. function function Petrovietnam co-operates with potential partners ANN has been developed to simulate X5 Σ the neural structure and activity of the Activation Transfer human brain. Deep learning (DL) is a function function Figure 1. Architecture of an ANN to manufacture polyester fiber ...... 69 Date of receipt: 21/8/2017. Date of review and editing: 21/8 - 5/9/2017. Date of approval: 4/10/2017.
14 PETROVIETNAM - JOURNAL VOL 10/2017 14 Petrovietnam and PVEP sign outsourcing contract for operation of Blocks 01 & 02 ...... 70 VPI signs co-operation agreement with Industrial University of Tyumen ...... 70 Amended gas sales and purchase and transportation contracts signed for Block 06-1, Nam Con Son basin ...... 71 Proposed measures to boost production in Bir Seba project, Algeria ...... 72 Phu My Fertilizer Plant meets target 2 months ahead of schedule ...... 72 KVT meets LPG and Bach Ho condensate production targets ahead of time ...... 73 Ca Mau Gas Processing Plant supplies first LPG shipment ...... 73
30 FOCUS
Prime Minister Nguyen Xuan Phuc: PETROVIETNAM TO STAND FIRM IN THE MIDST OF DIFFICULTIES
On 12 October 2017, at the Government’s headquarters, Prime Minister Nguyen Xuan Phuc worked with key leaders of the Vietnam Oil and Gas Group to evaluate the results of implementation of the production and business plan in 2017 and the deployment of key oil and gas projects.
4 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Prime Minister Nguyen Xuan Phuc works with key leaders of the Vietnam Oil and Gas Group. Photo: Quang Hieu/VGP t the meeting, the Prime of the country. Especially, the Prime In addition, Petrovietnam produced Minister confi rmed that Minister highly appreciated the great 8.25 billion m3 of gas, 17.09 billion Athe oil and gas sector contribution of Petrovietnam to the kWh of electricity, 1.48 million tons of plays an important role in socio- State Budget even in the diffi cult urea (exceeding the plan by 10.6%), economic development and has period. and 5.03 million tons of petroleum made huge contributions to the products (exceeding the plan by According to Petrovietnam’s national economy. 20.7%). President and CEO Nguyen Vu Truong Throughout the development Son, the Group’s production and With average oil price reaching process, the Party and the State business targets were all overfulfi lled USD 54.3/barrel, the total revenue of have given guidance by resolutions, from 2 - 21%. the Group amounted to VND 404.2 strategies, master plans and the Prime trillion, representing an increase of In the fi rst 10 months of 2017, Minister’s decisions to orientate the 15% compared to the 10-month oil production reached 13.01 million development of the Vietnam Oil and plan and 10% compared to the same tons, exceeding the 10-month plan Gas Group. period of 2016. Petrovietnam has by 2.8% and representing 85.6% contributed VND 76.1 trillion to the In fact, the Vietnam Oil and Gas of the yearly plan. The Group has State Budget for the fi rst 10 months Group has built a united team of produced 11.39 million tons of oil of 2017, which is 2% higher than the professional and highly qualifi ed in the country (exceeding the plan yearly plan. staff to overcome challenges and by 323 thousand tons, equivalent develop the oil and gas industry into to 3%); and 1.63 million tons of oil Also at the meeting, the Prime a key economic and technical sector abroad (exceeding the plan by 1.7%). Minister listened to the proposals and
PETROVIETNAM - JOURNAL VOL 10/2017 5 FOCUS
Prime Minister Nguyen Xuan Phuc highly appreciates the great contribution of the oil and gas industry to the State Budget even in the diffi cult period. Photo: Quang Hieu/VGP
PETROVIETNAM OVERFULFILS STATE BUDGET CONTRIBUTION PLAN FOR 2017 In the fi rst 10 months of 2017, oil production reached 13.01 million tons, exceeding the 10-month plan by 2.8% and representing 85.6% of the yearly plan. The Group has produced 11.39 million tons of oil in the country (exceeding the plan by 323 thousand tons, equivalent to 3%); and 1.63 million tons of oil abroad (exceeding the plan by 1.7%). In addition, Petrovietnam produced 8.25 billion m3 of gas, 17.09 billion kWh of electricity, 1.48 million tons of urea (exceeding the plan by 10.6%), and 5.03 million tons of petroleum products (exceeding the plan by 20.7%). With average oil price reaching USD 54.3/barrel, the total revenue of the Group amounted to VND 404.2 trillion, representing an increase of 15% compared to the 10-month plan and 10% compared to the same period of 2016. Petrovietnam has contributed VND 76.1 trillion to the State Budget for the fi rst 10 months of 2017, which is 2% higher than the yearly plan. recommendations of Petrovietnam’s Prime Minister Nguyen Xuan Prime Minister Nguyen Xuan key offi cials to solve diffi culties Phuc stressed: “We are responsible to Phuc requested the Vietnam Oil and obstacles so as to enable the the country and the people”, so each and Gas Group to concentrate all its continued development of the oil person, each unit must continue resources on completing the 2017 and gas industry in the coming to unite and strive to build on past targets; review and develop the period. In particular, the Prime achievements to contribute to the 2018 production and business plan Minister wanted Petrovietnam “to development of the Vietnam Oil and in line with the country’s growth stand fi rm in the midst of diffi culties”. Gas Group. objectives; and continue with the
6 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Bach Ho fi eld. Photo: Minh Tri equitisation process in accordance On behalf of more than 55,000 implement the tasks assigned by the with the plan approved by the employees of the oil and gas sector, Government, contributing to the stability and further development of Prime Minister. The Group focuses Petrovietnam’s President and CEO Nguyen Vu Truong Son confi rmed the Petrovietnam. on deploying key projects such as Nguyen Hoang that the Group will continue to Ca Voi Xanh, Block B, and Dung Quat uphold the will, the responsibility, Refi nery upgrading and expansion the competences and the tradition project. of the oil and gas staff to successfully
PETROVIETNAM - JOURNAL VOL 10/2017 7 FOCUS
DDeputyeputy PrimePrime MMinisterinister TTrinhrinh DinhDinh Dung:Dung: DUNG QUAT REFINERY IS THE LEVER OF QUANG NGAI’S ECONOMY On 19 October 2017, eporting to the Deputy capacity (106 - 108% on average). Prime Minister, General BSR has focused on implementing Deputy Prime Minister Trinh RDirector of BSR Tran solutions to optimise production Dinh Dung paid a working Ngoc Nguyen said Dung Quat operation; promoting scientifi c visit to Binh Son Refining and Refi nery has always been operating research, with primary emphasis Petrochemical Co. Ltd. (BSR). stably, reaching the optimal on energy savings, and optimising
8 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Dung Quat Refi nery. Photo: BSR the operational conditions of Budget, and earned an after-tax strategic investors to further invest in technological processes to reduce profi t equal to 341.7% of the yearly the development of petrochemical production costs and improve plan. and deep-processing sub-sectors in order to improve its production and production and business effi ciency. BSR informed that the company business effi ciency. In the fi rst 9 months of 2017, will launch the initial public off ering BSR produced 4.4 million tons of (IPO) in the 4th quarter of 2017 BSR will work on the investors’ products, and sold nearly 4.4 million according to the plan and expects proposals in detail for the purpose tons. BSR gained a total turnover of to sell only 5 - 6% of its shares. After of launching a successful IPO and VND 54.982 billion and contributed that, BSR will continue to seek and subsequent sale of shares to strategic VND 6,522 billion to the State select institutional investors and investors, especially Petrolimex, right
PETROVIETNAM - JOURNAL VOL 10/2017 9 FOCUS
Deputy Prime Minister Trinh Dinh Dung listens to reports on the progress of the Dung Quat Refi nery upgrading and expansion project. Photo: BSR after the plan is approved by the Deputy Prime Minister Trinh fl uctuations. At the same time, BSR Prime Minister. Dinh Dung affi rmed that the Dung needs to continue with product Quat Refi nery is the heart of Dung restructuring, investment, and With the aim of increasing Quat Economic Zone, which is the corporate governance to cut down the capacity and the fl exibility in economic lever of Quang Ngai expenditures on input and move the processing of crude oil, and province and the Central region of towards reduction of production upgrading technology to Vietnam. This is a demonstration that costs. ensure production of high- Vietnam can master the technology quality petroleum products and Emphasising the Government’s of oil refi ning and independently enhance competitiveness, BSR point of view that the State has the produce petroleum products to is implementing the project to responsibility to help businesses serve the economy. upgrade and expand Dung Quat operate eff ectively, the Deputy Refi nery to bring its capacity from Deputy Prime Minister Trinh Prime Minister affi rmed that the the current 6.5 million tons/year to Dinh Dung requested BSR to Government will support BSR to 8.5 million tons/year. The project has improve production capacity as complete its IPO as well as to upgrade been under way for 29/78 months. well as production and business and expand the Dung Quat Refi nery as scheduled. The contract for the EPC package is effi ciency in order to increase the Hong Minh expected to be signed in April 2018 competitiveness. BSR has been with the plant to be ready for start- effi cient in production but should up on 18 December 2021. not underestimate the market
10 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM PVEP and FPT co-operate in research and development of new technologies
PVEP and FPT sign a co-operation agreement on research and development of new technologies. Photo: PVEP
Petrovietnam Exploration nder this agreement, and select one of PVEP’s on-going FPT will provide support projects to pilot the application of Production Corporation Uto PVEP to digitalise new technological solutions. (PVEP) and FPT Joint Stock its operations in order to enhance FPT Chairman Truong Gia Binh Company (FPT) on 17 October the effi ciency of its oil and gas said that only industry 4.0 would 2017 signed a co-operation production, minimise downtime due help solve the specifi c problem in agreement on research to system incidents, save costs and the oil and gas sector which is how reduce risks for PVEP. and development of new to optimise production while still technologies in the fields of The two sides will promote ensuring the fi eld safety, cutting oil and gas exploration and long-term co-operation and apply costs of equipment investment and production. technological solutions on the basis shortening the exploration period. of the most advanced technologies The oil and gas sector with the such as internet of things (IoT), big application of high technologies and data, artifi cial intelligence (AI) and techniques has already been near the data science. In the coming period, optimum level, a few percentage of the two sides will meet, discuss improvement could bring signifi cant
PETROVIETNAM - JOURNAL VOL 10/2017 11 FOCUS
DR. NGO HUU HAI - PRESIDENT & CEO OF PVEP: The approach and application of 4.0 industrial technologies must be fast, neat and accurate to avoid falling behind. PVEP will pilot the application of new technologies, fi rst in wells with simple structure, then expand to other projects at home and abroad. economic benefi ts. He also asserted that in the oil and gas sector and software technology, the intellect and competences of Vietnamese people can compete in equal terms with other countries in the world. In the context where oil and gas companies are all facing diffi culties due to low oil prices, the application of advanced technological tools and solutions will facilitate more accurate data analysis, contribute to effi ciency improvement and cost optimisation in oil exploration and production.
Dr. Ngo Huu Hai, President & CEO technologies to reduce investment 6 million per day), we only need of PVEP said that as oil exploration costs and operating expenses, while an increment of 1% in output to always has risks, especially when oil increasing the continuous production decide on deployment”. Considering prices have remained at a low level time to 98 - 99%. for a long time together with other industry 4.0 as a very urgent matter, corollaries from previous unsuccessful “With the current scale of PVEP’s Dr. Ngo Huu Hai emphasised that projects, PVEP has to improve its operation (average production of the approach and application of 4.0 management level, promote the nearly 100 thousand barrels per industrial technologies must be fast, application of advanced and new day with estimated turnover of USD neat and accurate to avoid falling
12 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Dai Hung fi eld. Photo: PVEP behind. PVEP will pilot the application development for new technologies FPT will have the opportunity to of new technologies, fi rst in wells in oil exploration and production. access the specialised technical and with simple structure, then expand to PVEP will have an opportunity to gain technological environment of oil other projects at home and abroad. quick access to the latest trends of and gas exploration and production to test and improve the scope of By signing this agreement, industry 4.0, optimal solutions which applications, on that basis to better its are the strength of FPT, and then PVEP and FPT will strengthen services and technological solutions. their comprehensive and long- prepare resources to take advantage Manh Hoa term co-operation in research and of advanced technology. Whereas
PETROVIETNAM - JOURNAL VOL 10/2017 13 PETROLEUM EXPLORATION & PRODUCTION
APPLICATION OF DEEP LEARNING IN PREDICTING FRACTURE POROSITY Pham Huy Giao, Kushan Sandunil Geo-exploration & Petroleum Geoengineering Program, Asian Institute of Technology (AIT) Email: [email protected]
Summary
Deep learning (DL) neural network analysis is the latest development from the Artifi cial Neural Network (ANN) and it is being sedu more and more in petroleum engineering. In this study, the way to develop a new DL model for well log analysis was attempted and successfully implemented using well log data from a location in the Cuu Long basin, Vietnam. Three sets of analyses were conducted, i.e., the fi rst analysis set with a single hidden layer ANN model, the second analysis set with multiple hidden layer ANN model and the third with a DL neural network model. The DL-predicted porosity for a fractured granite basement reservoir of an oil fi eld in het Cuu Long basin was found in the range from 0.0 to 0.082, showing a good match with the conventionally-calculated values. The fi nal deeplearning model consists of 5-input layers of gamma ray (GR), deep resistivity (LLD), sonic (DT), density (RHOB) and neutron porosity (NPHI), having 5 hidden neuron layers with 14 neurons per layer. It is worth noting that the transfer function of the rectifi ed linear unit (ReLU), typical for a deep learning analysis, was implemented to replace the common sigmoidal transfer function, ensuring the successful application of DL model. Last but not least, the problem of vanishing gradient specifi c for a DL neural network model was also explained in details in this paper. Key words: Deep Learning (DL), Artifi cial Neural Network (ANN), fracture porosity, well log analysis, fractured granite basement, Cuu Long basin.
1. Introduction complex version of artifi cial neural networks. It is now widely used in image recognition, voice processing and language translation. The concept of soft computing was fi rst put forward by a paper named “Possibility Artifi cial neural networks are used to build a connection between theory and soft data analysis” [1]. While input and output data that helps predict the outputs of a set of new hard computing needs a precisely defi ned inputs (Figure 1). analytical model, soft computing is tolerant Application of deep learning in petrophysics is still in its infancy of uncertainty, imprecision, approximation stage. Classic neural networks commonly use one hidden layer, whereas and partial truth. This is a method designed deep learning uses multiple hidden layers. The use of multiple hidden to model and solve real world problems, layers can cause a phenomenon called vanishing gradient problem. which cannot be modelled mathematically. Deep learning network is obtained by eliminating this problem. The concept of soft computing was designed based on the concept of human Input layer Hidden layer brain functioning. X1 Σ Soft computing consists of few principal components as listed below [2]: X2 Σ - Fuzzy logic; Output layer X Σ Σ k - Evolutionary computation; 3 - Neural computing;
X4 Σ Activation Transfer - Probabilistic reasoning. function function
ANN has been developed to simulate X5 Σ the neural structure and activity of the Activation Transfer human brain. Deep learning (DL) is a function function Figure 1. Architecture of an ANN
Date of receipt: 21/8/2017. Date of review and editing: 21/8 - 5/9/2017. Date of approval: 4/10/2017.
14 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Conventional methods to estimate the primary obtained from core analysis data. Optimal porosity (or intergranular) porosity of clastic reservoirs are not prediction was given by an ANN model with one hidden helpful to estimate the fracture porosity. Therefore, layer having 15 neurons. Optimal permeability prediction petrophysicists keep trying to fi nd new techniques for was found by an ANN model having one hidden layer with fracture porosity estimation. 9 neurons. Nakapraves [8] carried out a research, using two ANN models to predict reservoir porosity for a clastic oil Kohli and Arora [3] did a research on application of fi eld in the northern Pattani basin, Gulf of Thailand. In the artifi cial neural networks for well log analysis. They used fi rst model, GR, LLD, RHOB, NPHI and DT logs were used as data from three wells including gamma ray (GR), resistivity input data. In the second model, the seismic attributes such (RES), density (RHOB), neutron porosity logs (NPHI) as the as dip, azimuth, instantaneous phase and relative acoustic input data and the core permeability as the target data for impedance were used as input parameters. Core porosity the ANN analysis. Kohli and Arora also concluded that ANN was used as the target values in both models. It was found could be relied upon for determination of characteristics that the ANN model trained using seismic attributes gave even in areas where cores are not available. The estimation far better results than the model trained using well log was completely data-driven and did not require any prior data. Foongthoncharoen [9] did a research on prediction of assumption. More than that the method was cost eff ective permeability and water saturation using neuron networks since it did not require labourious core analysis. and fuzzy logic for a clastic reservoir in the Gulf of Thailand. Korjani et al. [4] carried out a research on a new Permeability and water saturation were predicted for two approach of reservoir characterisation using deep wells. Well log data of gamma ray, deep resistivity, medium learning neural networks for Kern River fi eld located in resistivity, density and neutron porosity were used as San Joaquin valley, California. The fi eld consists of nine input parameters. Core permeability and water saturation production formations. The study was carried out using calculated by using Indonesian model were used as target data from 473 wells. Well log data such as deep resistivity, values. Between the two models of ANN and fuzzy logic, medium resistivity, and neutron porosity were used as the latter one gave better results. Sakulluangaram [10] did input parameters. The authors came to conclude that a research on petrophysical modelling for the Miocene deep learning is an eff ective method for characterisation Sand in the South Fang basin, northern Thailand, where of a reservoir with large volumes of data. he predicted porosity distribution of a well (well A) using the data from a nearby well (well B). GR, RHOB, NPHI and Guler et al. [5] carried out a study on predicting DT logs were used as the input parameters. Porosity values relative permeability of a hydrocarbon reservoir using an predicted were similar to the porosity obtained from well artifi cial neural network. First, they found what should be logs. Thanh [11] did an ANN prediction of porosity using the key input data, i.e., common fl uid and rock properties the integrated well log and seismic attribute data including were selected as the input data set. In addition, when refl ection amplitude, instantaneous amplitude, its 1st and selecting these data, they focused on parameters that 2nd derivative, as well as instantaneous phase, frequency, could be measured in laboratories and/or easily obtained dominant frequency and refl ectivity as the input for from literature. Based on the results of their study, the an ANN model. The fi nal porosity output determined authors concluded that with the increase in the number using a Bayesian Regulation training algorithm gave the of property-based input parameters, the effi ciency of the best match with the highest R-value of 0.86. Kano [12] ANN model was enhanced. predicted porosity for a fractured basement reservoir In the last decade, researchers from the Asian Institute with well log data from two wells using a conventional of Technology (AIT) had conducted a number of studies method proposed by Elkewidy and Tiab [13] and fi ve soft on application of soft computing in well log analysis computing techniques, including ANN, Fuzzy Inference using fuzzy logic and neural computing [6]. For example, System with Mamdani’s style, Fuzzy Inference System with Witthayapradit [7] carried out a research on an integrated Sugeno’s style, Fuzzy Subtractive Clustering and Adaptive petrophysical study using well logging data for enhancing Neuro-Fuzzy Inference System (ANFIS). These Artifi cial a gas fi eld in the Gulf of Thailand. He originally used gamma Neural Networks gave the best prediction. Duangngern ray (GR), deep resistivity (LLD), density (RHOB), neutron [14] did a fuzzy analysis to predict the eff ective porosity of porosity (NPHI) and sonic log (DT) as the input parameters a clastic reservoir in the Pattani basin, Thailand. He applied to predict porosity and permeability. Target values were various Fuzzy logic techniques, using well logging and
PETROVIETNAM - JOURNAL VOL 10/2017 15 PETROLEUM EXPLORATION & PRODUCTION
core data of two wells. Among three fuzzy techniques Deep learning is a machine learning algorithm which used, the fuzzy subtractive clustering was found the best could learn complex functions. The learning process of for prediction of eff ective porosity with low value of root deep learning algorithm fi lters important information mean square error, high value of R2 and R-value. Wang from raw data in a systematic way. [15] carried out a research on ANN-based prediction of Deep learning architecture consists of multiple fractured rock mass hydraulic conductivity for the Frieda hidden layers comparing to shallow neural networks River copper-gold mine in Papua New Guinea. In this that used to have one hidden layer. With multiple hidden study, the backpropagation neural network (BPNN) was layers, a problem called vanishing gradient occurs during trained to successfully map the relationship between back propagation, thus any deep learning network should indicative rock parameters and hydraulic conductivity be able to overcome this problem. using a variety of rock data sets collected in the site Deep learning networks, like other ANN’s, have a investigation of a feasibility study for the study mine. Input predicted and an expected (target) value. The intention data were selected for ANN analysis included lithology, of learning is to make the diff erence between the target weathering, fracturing, unconfi ned compressive strength and the expected values as small as possible (which (UCS), defect angle and rock quality designation (RQD). we assign as the goal). To understand how vanishing Diff erent numbers of input parameters (4, 5 and 6) were gradient occurs, a simple deep neural network to predict used to build the ANN model. Wang [15] also investigated a cost function with single neuron in each hidden layer is the eff ect of various transfer functions on the ANN model considered as shown in Figure 2, where x is the input, W , by using log-sigmoid, tan-sigmoid and purelin. 1 W2, W3, and W4 are the weights, b1, b2, b3, and b4 are the 2. Vanishing gradient problem of multiple deep biases, and J is the cost function of the network. learning neural network analysis In Figure 3, J represents the cost function, which is a function of the diff erence between predicted and W W W W expected values. Artifi cial neural networks usually make X 1 b 2 b 3 b 4 b J 1 2 3 4 use of the following sigmoid function as the activation Figure 2. Simple multi-layer ANN model after introducing the cost function function or transfer function: (1) Where: S(n): The sigmoid function n is variable One of the reasons for popular use of the sigmoid function is that its derivative can be easily calculated by the very value of sigmoid function as follows: ‘ (2) Where: Figure 3. Graphical representation of sigmoid function [12] S' (n): The derivative of sigmoid function
W Vanishing gradient problem 1 (a) X b1 J
One hidden layer b 1 W W 1 2 b (b) X b1 2 J
Two hidden layers W W W 1 2 b 3 X b1 2 b3 J (c)
Three hidden layers Figure 4. Diff erent ANN modes: (a) ANN model with one hidden layer; (b) ANN model with two hidden layers; and (c) ANN model with three hidden layers Figure 5. Graphical representation of the derivative of sigmoid function [12]
16 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
The derivative of sigmoidal function is plotted in Where: Figure 5 which shows it lies between 0 and 0.25. S: Sigmoid function; By the chain rule, the derivative of the error (J) with z2: Input to the second hidden layer; respect to the fi rst weight (W1), can be written as: z3: Input to the third hidden layer; (3a) z4: Input to the output layer.
(3b) (5a)
(3c) (5b) (3d)
(3e) (5c) Where: The maximum value that the derivative of the aout: Output of the ANN network; sigmoid function can take is 0.25 (and the minimum is 0). a : Output from the fi rst hidden layer; 1 The weights are selected based on Gaussian distribution,
a2: Output from the second hidden layer; thus the value of weights always lies between -1 and 1.
a3: Output from the third hidden layer; When the values of each derivative are multiplied by each other, the resultant will give a small value less W : Weights of the input layer; 1 than 1. Imagine if the number of hidden layers increased,
W2: Weights of the fi rst hidden layer; becoming deeper and deeper, the number of derivative terms will increase as well as the number of terms to be W3: Weights of the second hidden layer; multiplied (Equations 5a-c), making the resulting value of W : Weights of the third hidden layer; 4 smaller and smaller. This shows how early layers x: Inputs of the ANN; learn slower with the increase of the number of hidden S: Sigmoid function. layers. A similar approach can be implemented to show Considering the following derivative components in that latter layers learn faster by obtaining the derivation of cost function with respect to W . Equation 3a 3 Vanishing gradient occurs because of using the sigmoid function. To overcome the problem, an alternative activation function known as rectifi ed linear unit (ReLU) was introduced in deep learning. In Figure 2, by denoting the inputs to the 4th, 3rd, 2nd hidden layers as z4, z3, z2 respectively, one can write the 3. Methodology of this study following relationships: The well log data were collected from a well drilled (4a) into a fractured granite basement (FGB) reservoir in an oil (4b) fi eld of Cuu Long basin for a depth range from 2,525m to 3,015m including gamma ray (GR), deep resistivity (LLD), (4c) shallow resistivity (LLS), interval transit time (DT), bulk density (RHOB), neutron porosity (NPHI), photoelectric (4d) factor (PEF), and caliper (CAL). Figure 6 shows the general (4e) workfl ow of this study.
(4f) The total and fracture porosity for the target data set was calculated using the approach suggested by Elkewidy
PETROVIETNAM - JOURNAL VOL 10/2017 17 PETROLEUM EXPLORATION & PRODUCTION
The formation factor (F): R a Well log data collec on Deep learning model with 5 F (7a) R (GR, LLD, LLS, RHOB, DT, input parameters and 150 NPHI, PEF, CAL) training examples having rec fied linear unit (ReLU) and fracture intensity index (FII): as the transfer func on (7b)
Single hidden layer model By rearranging Equation 7b, the matrix porosity is: with 5 input parameters and (7c) 150 training examples Single hidden layer, having log-sigmoid as the mul ple hidden layer and As fracture porosity ( f) is: transfer func on deep learning models for 500, 1000, 2000 and 3000 (7d) training examples by replacing m in 7d with 7c, one has: (7e) Using different combina ons of input parameters on the For a double porosity fractured reservoir with single layer model to find the 100% formation fl uid saturation one can write: best input Results comparision (7f)
Where Rma is the resistivity of 100% fl uid
saturated matrix and Ro is the resistivity of the total Mul ple hidden layer model system (matrix voids + fracture voids) being 100% with 5, 6, 7, 8 input fl uid saturated. parameters with 150 training Rearranging Equation 7f and combining with examples having log-sigmoid as the transfer func on Equation 7a one gets:
(7g) Figure 6. Flow chart of the study’s methodology and Tiab [13] as shown in Equations 6a and 6b below. Matrix (7h) density was assumed as 2.71g/cc since formation density and porosity were measured in limestone scale. Assuming a = 1 and with further rearranging Equation 7h one has: (6a)
(7i) (6b) In case, one can neglect water resistivity (R ) in Where: w Equation 7i since Rma >> Rw, the porosity partitioning coeffi cient can be expressed as: Øt: Total porosity (fraction); NPHI : Neutron porosity (fraction); (7j) By replacing 7j with 7b one has: PHID: Porosity calculated from bulk density (fraction); (7k) RHOB: Bulk density of formation (g/cc); Further replacing Equation 7k with 7c and 7e one
ρma: Matrix density, assumed to be 2.71g/cc (since gets: measured in limestone scale); (7l) ρ : Fluid density, assumed to be 1g/cc (water). f Hence the target (fracture porosity) will be: An equation to calculate fracture porosity of a fractured (7m) reservoir can be derived by using the following steps:
18 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Where: method proposed by Elkewidy and Tiab [13] as explained above in detail (Equation 7f). Øf: Fracture porosity (fraction); 4. Results and discussion Øma: Matrix porosity (fraction); m: Cementation exponent. In Analysis I, ANN models with a single hidden layer were analysed. Firstly, 16 models numbered from MI1 to MI16 (Table 1) were tested with Partitioning coeffi cient (ν), matrix 150 training examples to fi nd out the best combination of fi ve input porosity (Ø ), and fracture porosity (Ø ) ma f parameters out of 8 well log data used (GR, LLD, LLS, DT, RHOB, NPHI, PEF are calculated by Equations 7j, 7l and 7m, and CAL). It was found that the model with the basic 5-input sets (GR, LLD, respectively [13]: RHOB, NPHI and DT), 12 neurons of the hidden layer and running for 150 Data from the depth interval of 2,515 training examples (Figure 7) gave the best prediction of fracture porosity - 3,015m were selected and analysed for with the least prediction error (Table 2). Four more models in Analysis each 0.125m interval. Three conducted I, numbered from MI17 to 20 (Table 1), were tested with the number analyses are summarised in Table 1 and of training examples increased from 150 to 500; 1,000; 2,000 and 3,000 named as I, II and III. In analysis I, only respectively, and with diff erent number of neurons in the hidden layer. fi ve inputs out of the eight well log However, prediction of fracture porosity did not improve (Figure 7). parameters were investigated. A total of In Analysis II, various multiple hidden layer ANN models (Table 1) 16 combinations, denoted as MI1 to 16 were tested to see if the prediction of fracture porosity would increase (Table 1), will be done to fi nd out the best with the increasing number of hidden layers using the basic set of fi ve combination of 5. Based on the criteria proposed by Guller et al. [5], the combination of fi ve selected input parameters having the least prediction error at 150 training examples will be further tested for training examples of 500; 1,000; 2,000 and 3,000, respectively. In analysis II, with multiple hidden layer ANN model, the fi rst issue to be studied is what would happen if one uses more than 5 input parameters, for example 6, 7 and 8 input parameters. After that, the eff ect of the number of training examples will be also Figure 7. Fracture porosity calculated by single hidden layer ANN models, Analysis I, with 150; 500; 1,000; 2,000 investigated. In analysis III, a deep learning and 3,000 training examples as seen from left to right model will be developed from the best multiple hidden layer model found in Analysis II by replacing the sigmoidal transfer functions with the rectifi ed linear units (RELU). The deep learning (DL) models will also be investigated for various number of hidden layers as well as number of training examples, i.e., 50; 500; 1,000; 2,000 and 3,000. The fracture porosity values predicted by all of the ANN and DL models in analyses I, II and III will be plotted and compared between them as well with the Figure 8. Fracture porosity calculated by multiple hidden layer ANN models, Analysis II, with 150; 500; 1,000; fracture porosity values calculated by the 2,000 and 3,000 training examples as seen from left to right
PETROVIETNAM - JOURNAL VOL 10/2017 19 PETROLEUM EXPLORATION & PRODUCTION
input parameters as found in Analysis I. It was found that model MII1 In Analysis III, deep learning models with 4 hidden layers and 12 neurons per layer gave the least prediction were developed using the ReLU transfer error of 16.46% after 150 training examples. When the number of training function and tested. A total of fi ve models examples increased from 150 to 500; 1,000; 2,000 and 3,000 respectively, numbered from MIII1 - MIII5 as seen in the prediction of fracture porosity was not improved either (Figure 8). Table 1 were run with the basic set of fi ve input parameters (GR, LLD, RHOB, NPHI and DT) and with training examples of 150; 500; 1,000; 2,000 and 3,000. The MIII5 model was found as the best with 5 hidden layers and 14 neurons per layer for the case of 3,000 training examples. The deep learning predicted fracture porosity in the range from 0 - 0.082, which matched with those calculated by the conventional method by Elkewidy and Tiab [13]. 5. Some concluding remarks Figure 9. Fracture porosity calculated by deep learning models, Analysis III, with 150; 500; 1,000; 2,000 and 3,000 training examples In this study, a three-step approach Table 1. Summary of ANN and deep learning models in this study Type of Neural Number of training Input Parameters Transfer function Networks (NN) examples Analysis Model Model Single Multiple Modifi Increased Deep Basic Rectified Remarks No. No. Code hidden hidden ed number 150 500 1,000 2,000 3,000 Log- learning five Linear layer layer five of inputs sigmoid (DLNN) inputs Unit (ANN) (ANN) inputs 6 7 8
Five basic inputs 1 MI1 V V V V GR, LLD, RHOB,
NPHI, DT MI2- Each parameter 2-6 V V V V MI6 replaced by a CAL MI7- Each parameter 7-11 V V V V I MI11 replaced by a LLS MI12- Each parameter 12-16 V V V V MI16 replaced by a PEF 17 MI17 V V V V Five basic inputs
18 MI18 V V V V Five basic inputs 19 MI19 V V V V Five basic inputs 20 MI20 V V V V Five basic inputs 21 MII1 V V V Five basic inputs
CAL, LLS, and PEF MII2- 22-24 V V V V added to basic 5- MII4 inputs CAL and LLS, CAL MII5- and PEF and LLS 25-27 V V V V MII7 and CAL added to II basic 5-inputs All 8 parameters 28 MII8 V V V V used 29 MII9 V V V Five basic inputs
30 MII10 V V V Five basic inputs 31 MII11 V V V Five basic inputs
32 MII12 V V V Five basic inputs 33 MIII1 V V V Five basic inputs
34 MIII2 V V V Five basic inputs
III 35 MIII3 V V V Five basicinputs 36 MIII4 V V V Five basic inputs
37 MIII5 V V V Five basic inputs
20 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Table 2. Average prediction error for diff erent number of neurons in single hidden layer ANN models, Analysis I, with 150; 500; 1,000; ,0002 and 3,000 training examples Neurons No. Number of hidden layers Model Model in the Training No. code hidden examples 3 4 5 6 7 8 9 10 11 12 13 14 15 layer 150 1 MI1 13 36 17 20 20 21 19 21 17 10 32 20 20 12 500 17 MI17 1,284 1,305 817 1,665 1,345 2,504 2,326 2,301 2,123 1,568 1,996 4,168 1,973 5 1,000 18 MI18 1,152 799 879 702 932 876 635 1,088 870 738 2,914 909 1,713 9 2,000 19 MI19 1,302 20,642 1,284 5,767 7,078 5,782 5,622 5,203 3,240 22,832 3,972 4,025 16,131 5 3,000 20 MI20 1,050 2,173 694 712 3,360 1,576 1,476 2,186 9,003 6,397 1,655 818 4,537 14
Table 3. Average prediction error for diff erent number of neurons in multiple hidden layer ANN models, Analysis II, with 150; 500; 1,000; 2,000 and 3,000 training examples Number of hidden layers Neurons No. Model Model Hidden in the Training No. code layers hidden examples 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 layer 150 21 MII1 19 22 16 34 22 31 18 47 29 38 23 33 40 23 48 31 44 70 66 4 12 500 29 MII9 1,503 3,355 2,313 1,882 1,499 1,222 452 994 3,123 2,842 2,079 8,410 595 1,927 475 1,106 1,752 450 1,236 19 5 1,000 30 MII10 1,045 948 477 841 735 559 566 605 493 267 545 480 556 534 513 380 627 1,012 356 11 9 2,000 31 MII11 1,045 4,016 462 961 46,535 5,058 3,127 803 4,983 2,427 3,512 606 1,683 2,691 2,307 1,530 4,548 3,558 833 4 5 3,000 32 MII12 3,804 882 405 387 2,717 2,197 1,503 1,223 429 1,304 1,691 290 417 1,244 762 700 573 1,280 1,315 5 14
Table 4. Average prediction error for diff erent number of neurons in deep learning model References with 150; 500; 1,000; 2,000 and 3,000 training examples Number of Number of 1. Lotfi A.Zadeh. Fuzzy logic, neural networks, and Model Training neurons per hidden soft computing. Communications of the ACM. 1994; 37(3): code examples hidden layer layers p. 77 - 84. MIII1 150 12 4 2. R.C.Chakraboty. Fundamentals of neural network. MIII2 500 5 19 Artifi cial Intelligence. 2010. MIII3 1,000 9 11 3. A.Kohli, P.Arora. Application of artifi cial neural MIII4 2,000 5 4 networks for well logs. International Petroleum Technology MIII5 3,000 14 5 Conference. 19 - 22 January, 2014. 4. M.Korjani, Andrei Popa, Eli Grijalva, Steve Cassidy, was successfully conducted to develop deep learning I.Ershaghi. A new approach to reservoir characterization models to predict the fracture porosity of a fractured using deep learning neural networks. SPE Western Regional granite basement (FGB) reservoir in the Cuu Long basin Meeting, Anchorage, Alaska, USA. 23 - 26 May 2016. for the depth interval from 2,515m to 3,015m. The facture porosity was found in a range from 0.0 to 0.084, which 5. B.Guler, T.Ertekin, A.S.Grader. An artifi cial neural is matching quite well with the values calculated using network based relative permeability predictor. Journal of the conventional method by Elkewidy and Tiab [13]. The Canadian Petroleum Technology. 2003; 42(4). results of this study showed that for the large volumes 6. P.H.Giao. Application of ANN in petrophysics. of well data (number of training examples) the more Lecture notes, Asian Institute of Technology, Bangkok, traditional ANN models with a single hidden layer could Thailand. 2008. not work well, but the DL model could. Thus, to apply 7. T.Witthayapradit. Formation evaluation using more eff ectively the neural network analyses in analysis of integrated well logging data for a gas fi eld in the Gulf of well log data at the industrial scale, one may try to employ Thailand. Master thesis, Asian Institute of Technology, the DL models with multiple hidden neuron layers and Bangkok, Thailand. 2009. ReLU transfer function instead of one-hidden layer ANN models. 8. N.Nakapraves. Application of ANN analysis in prediction of reservoir porosity for an oil fi eld in the Northern
PETROVIETNAM - JOURNAL VOL 10/2017 21 PETROLEUM EXPLORATION & PRODUCTION
Pattani basin. Master thesis, Asian Institute of Technology, thesis, Asian Institute of Technology, Bangkok, Thailand. Bangkok, Thailand. 2011. 2014. 9. T.Foongthoncharoen. Prediction of permeability 13. T.I.Elkewidy, D.Tiab. An application of conventional and water saturation using neuron networks and fuzzy well logs to characterize naturally fractured reservoirs with logics for a clastic reservoir in the Gulf of Thailand. Master their hydraulic (fl ow) units; a novel approach. SPE Gas thesis, Asian Institute of Technology, Bangkok, Thailand. Technology Symposium, Calgary, Canada. 1998. 2012. 14. D.Duangngern. Application of Fuzzy Analysis to 10. C.Sakulluangaram. Petrophysical modelling for the predict the eff ective porosity in a clastic reservoir, Pattani Miocene sand in the South Fang basin. Master thesis, Asian basin. Master thesis, Asian Institute of Technology, Institute of Technology, Bangkok, Thailand. 2013. Bangkok, Thailand. 2015. 11. D.V.Thanh. Prediction of porosity by ANN analysis 15. Y.Wang. ANN-based prediction of fractured rock integrating well log and seismic attribute data for an oil mass hydraulic conductivity for the Frieda river copper-gold fi eld in the Cuu Long basin, off shore Vietnam. Professional mine in Papua New Guinea. Master thesis, Asian Institute of Master Research Study, Asian Institute of Technology, Technology, Bangkok, Thailand. 2016. Bangkok, Thailand. 2013. 16. R.Kapur. The vanishing gradient problem. A Year 12. N.Kano. Soft computing - Based prediction of of Artifi cial Intelligence. 2016. porosity for a fractured granite basement reservoir. Master
22 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
DIFFERENT FORMS OF GASSMANN EQUATION AND IMPLICATIONS FOR THE ESTIMATION OF ROCK PROPERTIES Tran Trung Dung1, Carl H.Sondergeld2, Jean-Claude Roegiers2 1Phu Quoc POC 2University of Oklahoma Email: [email protected]
Summary
Three new equivalent forms of Gassmann equation are presented that are useful when the unknown parameters are the Biot- Willis coeffi cient, the dry bulk modulus, and/or the grain matrix bulk modulus. We apply these equations to several sets of laboratory measurements to determine the profi les of grain matrix bulk modulus and Biot-Willis coeffi cient as functions of applied pressure, and perform a Monte Carlo simulation to examine the eff ect of uncertainty and/or measurement errors on the calculated grain matrix bulk modulus and Biot-Willis coeffi cient. The results show that the calculated grain matrix bulk modulus is relatively constant with applied diff erential pressure (up to 50MPa). However, it is very sensitive to the uncertainty of dry and saturated bulk modulus values. Thus, the presented new forms of Gassmann equation can be used to eff ectively quantify the uncertainty of dry and saturated bulk modulus (and subsequently, the seismic velocities) in fl uid identifi cation, fl uid substitution, or reservoir monitoring applications. Key words: Gassmann equation, bulk moduli, Biot-Willis coeffi cient, sensitivity analysis.
1. Introduction Berryman [5] gave a concise derivation of Gassmann equations for an isotropic and homogeneous medium The Gassmann equations [1] have been used using the quasi-static poroelastic theory. Other extensively in the oil and gas industry for fl uid identifi cation forms of Equation (1) can be found in Mavko et al. [6]. and reservoir monitoring applications, despite its various Zimmerman [7] presented an equivalent form in terms assumptions [2, 3]. The fi rst Gassmann equation provides of compressibility. However, Equation (1) is probably the the relationship between the saturated bulk modulus most intuitive in describing the eff ect of fl uid presence on of a rock and its dry frame bulk modulus, porosity, bulk the bulk modulus. modulus of the mineral matrix, and bulk modulus of the pore-fi lling fl uid. Whereas, the second Gassmann White and Castagna [8] argued that, since all input equation simply states that the shear modulus of the rock parameters for Gassmann equations carry some degrees is independent of the presence of the saturating fl uid: of uncertainty, a fl uid modulus inversion should be performed using a probabilistic approach. Artola and K α 2 f Alvarado [9] evaluated the eff ect of uncertainty of K = K + (1) sat dry K diff erent input parameters and showed that the computed f φ + ()α − φ compressional velocity of a saturated rock is most sensitive K m to uncertainties in the rock bulk density and the dry bulk G=G sat dry (2) and shear moduli, while other parameters (porosity, the grain matrix and fl uid bulk moduli) have negligible eff ects. Where α is the Biot-Willis coeffi cient [4]: K Note that the three parameters: dry frame modulus dry (3) α =1− (Kdry), Biot-Willis coeffi cient (α), and grain matrix bulk K m modulus (Km) are related by Equation (3); in many instances The moduli are related to the seismic velocities and they are unknown. The fl uid saturated bulk modulus (Ksat) density by: and fl uid bulk modulus (Kf) can also be unknown (e.g. in ⎛ 2 4 2 ⎞ fl uid substitution problem). As a result, ad-hoc and empirical K = ρ ⎜V − Vs ⎟ (4) ⎝ p 3 ⎠ correlations have been proposed to address this problem. There are many instances Biot-Willis coeffi cient is assumed G = ρV 2 s (5) to be 1 due to the lack of a better estimate. For sandstone
Date of receipt: 20/9/2017. Date of review and editing: 20/9 - 5/10/2017. Date of approval: 5/10/2017.
PETROVIETNAM - JOURNAL VOL 10/2017 23 PETROLEUM EXPLORATION & PRODUCTION
at high diff erential pressure (40MPa), Han and Batzle [10] Therefore, instead of having two non-linear equations proposed α to be a polynomial function of porosity: for two unknowns (α and Km), we reduce the problem to α = 3.206φ − 3.349φ 2 + 1.143φ 3 (6) one simple quadratic equation, Equation (7), that always gives one physically realistic solution. In this study, we present 3 new equivalent forms of This provides an independent methodology to Gassmann equation that are useful for diff erent scenarios calculate the grain matrix bulk modulus of a rock from SCAL of available data. We apply these equations to several laboratory acoustic measurements of dry and saturated sets of laboratory measurements. A stochastic simulation rock samples. Traditionally, the grain matrix bulk moduli is performed to examine the eff ect of uncertainty and/ are estimated from averages of the rock mineralogical or measurement errors on calculated grain matrix bulk composition (e.g. Voigt-Reuss-Hill average or Hashin- modulus and Biot-Willis coeffi cient. Shtrikman bounds). These bounds may carry large 2. The new equivalent Gassmann equations uncertainties since many minerals, especially clays, have a high variance in their bulk modulus values depending 2.1. When (K , K , K , and φ) are known dry sat f on the measurement conditions [12, 13]. We can further This is generally the case for laboratory measurements postulate that: (a) the grain matrix calculated from on dry and wet rock samples (e.g. dry and brine saturated Gassmann equation (using Equations (8) to (12)) must lie acoustic velocities are measured as functions of diff erential between the two bounds obtained from mixture theory, pressure along with rock porosity). In this case we can and (b) the calculated grain matrix values are insensitive rewrite Equation (1) as a function of Biot-Willis coeffi cient to the fi rst order to the applied pressure. Equations (7) to α (see Appendix A for a detailed derivation): (11) can also be used to verify the applicability of existing empirical or ad-hoc correlations (such as Equation (6) to ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ 2 dry dry ⎜ dry ⎟ α − (φ + 1)⎜1− ⎟α +φ⎜1− ⎟ 1− = 0 (7) estimate Biot-Willis coeffi cient) for diff erent rocks. ⎜ K ⎟ ⎜ K ⎟⎜ K ⎟ ⎝ sat ⎠ ⎝ sat ⎠⎝ f ⎠ 2.2. When (Ksat1, Ksat2, Kf1, Kf2, and φ) are known Equation (7) is a quadratic equation A2 + B + C = 0, This case can be encountered in the fi eld. The same where all coeffi cients can be readily calculated. rock can be fully saturated with brine in one well while A = 1 (8) having oil or gas in another well; or it can have varying ⎛ K ⎞ dry saturations in the same well. Acoustic logs, and density B1= −(φ + 1)⎜ − ⎟ (9) ⎜ K ⎟ - porosity logs are available. In this case, Kdry, Km, and α ⎝ sat ⎠ are unknown in a system of three non-linear equations ⎛ K ⎞⎛ K ⎞ dry dry (two Equations (1) for two diff erent saturation fl uids and C1= φ ⎜1− ⎟⎜ − ⎟ (10) ⎜ ⎟⎜ ⎟ Equation (3)). Starting from Equation (7) instead, we end Ksat K f ⎝ ⎠⎝ ⎠ up with (see Appendix B for detailed derivations): This simple quadratic equation has two solutions: ⎛ 1 1 ⎞ ⎛ 1 1 ⎞ φ ⎜ − ⎟K =φ⎜ − ⎟ − − B ± ∆ 2 ⎜ K K K K ⎟ dry ⎜ K K ⎟ α = , where ∆ = B4− AC (11) ⎝ sat 1 f1 sat2 f 2 ⎠ ⎝ f1 f 2 ⎠ 1,2 2A (13) ⎛ 1 1 ⎞ ⎜ ⎟ However, Berryman and Milton [11] showed that α is []α ()φ+1 −φ ⎜ − ⎟ ⎝ Ksat1 Ksat2 ⎠ physically bounded between 0 and 1. Equations (9) and We can write Equation (13) in a more convenient form (10) show that B is negative since Kdry < Ksat, and C is also for numerical calculations: negative since Kf < Kdry for consolidated rocks. This means − B + ∆ Δ = B2 - 4AC > B2 and thus, α = is the only 1 2A ⎛ K K ⎞ ⎛ ⎞ ⎜ sat2 sat1 ⎟ ⎜ 1 1 ⎟ possible solution since α is negative. φ − K = φ − K K 2 ⎜ K K ⎟ dry ⎜ K K ⎟ sat1 sat2 ⎝ f 1 f 2 ⎠ ⎝ f 1 f 2 ⎠ The corresponding grain matrix bulk modulus then (14) can be calculated from Equation (3): + ()K − K []α (φ+ 1) − φ K sat1 sat2 dry Km = (12) 1− α Kdry, α, and Km can now be calculated very quickly
24 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
using a simple iteration using Equation (14) and Equation And applying this α to Equation (3) gives Kdry. This is
(7) as follows: equivalent to the Kdry solution of Zhu and McMechan [14]. - Step 1: Make an initial guess, for example: 3. Numerical applications K×dry = 0.5 min{}Ksat1, Ksat2 3.1. Han and Batzle’s sandstone data - Step 2: Use guessed Kdry value in Equation (7) to fi nd two Biot-Willis coeffi cients αf1 and αf2 (for two saturations): We applied Equation (7) to the pressure dependent dry and brine saturated velocities and moduli of a porous ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ α 2 − (φ +1)⎜1− dry ⎟α +φ⎜1− dry ⎟⎜1− dry⎟ = 0 sandstone sample published by Han and Batzle [10] (Figure ⎜ K ⎟ ⎜ K ⎟⎜ K ⎟ ⎝ sat1⎠ ⎝ sat1⎠⎝ f1 ⎠ 1). The wet and dry densities are back-calculated from ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ Equations (4) and (5). The porosity values are calculated α 2 − (φ+1)⎜1− dry ⎟α +φ ⎜1− dry ⎟⎜1− dry ⎟ = 0 ⎜ K ⎟ ⎜ K ⎟⎜ K ⎟ using the density relationship: ⎝ sat2 ⎠ ⎝ sat2 ⎠⎝ f2 ⎠ ρ = ρ + φρ sat dry f (16) - Step 3: Take the average for a new Biot-Willis coeffi cient: The calculated Biot-Willis coeffi cient and grain matrix modulus as functions of pressure are plotted in Figure 1. α = ()α + α /2 f1f2 The relatively constant value of the grain bulk modulus - Step 4: Use this new α in Equation (14) to fi nd new (39GPa) as a function of pressure is a good indicator that Gassmann equation is applicable for this rock. The Kdry. variation of grain bulk modulus at low confi ning pressure - Step 5: Repeat steps 2 to 4 until K converges: dry (< 10MPa) is possibly due to higher uncertainty in input K − K values (i.e. higher noise-to-signal ratio from velocity dry,new dry,old < ε signals). Kdry,new The Biot-Willis coeffi cient profi le is remarkably similar - Step 6: Use Equations (7) and (12) to fi nd to the result measured on a 26% porosity Boise sandstone corresponding α and K . m sample by Fatt [15]. Note that Gassmann equation gave Note that we have assumed there are no softening a higher value (0.73 at 40MPa) than Han and Batzle’s or hardening eff ects caused by the saturating fl uids on Equation (6) (0.63). the grain bulk modulus (K is constant). The second m 3.2. Coyner’s limestone data assumption is that the rock dry frame is stiff er than both fl uids, K>dry max {}K f 1, Kf 2 , so that Equation We employed the iteration procedure using Equations (7) still gives only one positive (physically realistic) root. (7) to (14) on water and benzene saturated Bedford This assumption is generally valid for consolidated limestone sample published by Coyner [16] (Figure 2). sedimentary rocks. In his experiment at room temperature, the fl uid pore pressure in both saturation cases was maintained at 2.3. When (K , K , K , and φ) are known m sat f 10MPa. The porosity of the rock is 11.9%. The shear modulus profi le is almost identical for all vacuum dry, In this case K and α are unknown. An instance for dry water saturated, and benzene saturated cases, suggesting this case is that fl uid data, fl uid saturation, acoustic log that Gassmann equation is valid for this rock. At 10MPa (V , V ) and density-porosity log are available while K is p s m pore pressure, K = 2.24GPa, and K = 1.21GPa. estimated from the mineralogical composition of the rock water benzene (FTIR, XRD, thin section of rock cuttings, or mineralogy The back-calculated dry bulk modulus is also log). The Biot-Willis coeffi cient can be estimated directly plotted (as a dashed line) against the various measured from the following equivalent Gassmann equation (see moduli in Figure 2. The profi le is consistently higher Appendix C for detailed derivations): than the measured vacuum dry bulk modulus profi le by approximately 2.5GPa (or 5 - 9%). This is another ⎡ ⎛ K ⎞⎤ ⎡ ⎛ K ⎞⎤ ⎢φ()K1−K −K ⎜1− sat ⎟⎥α =φ ⎢()K −K −K ⎜ − f ⎟⎥ evidence supporting the argument that the vacuum m f f ⎜ K ⎟ m f sat ⎜ K ⎟ (15) ⎣ ⎝ m ⎠⎦ ⎣ ⎝ m ⎠⎦ dry measured bulk modulus is too dry and should not
PETROVIETNAM - JOURNAL VOL 10/2017 25 PETROLEUM EXPLORATION & PRODUCTION
be used in Gassmann equation [6, 17]. We could have 80 1.0 applied the measured vacuum dry and either water- or benzene-saturated moduli values on Equation (7), but 70 0.8 α that approach would give unrealistically high grain matrix 60 bulk modulus. 0.6 50 The grain matrix bulk modulus and Biot-Willis 0.4 coeffi cient profi les obtained from the rock water- and
40 coefficient -Willis benzene-saturated moduli are shown in Figure 3. While the Biot
Grain bulk modulus, GPa 0.2 30 Grain bulk modulus K m grain matrix bulk modulus is similar to Coyner’s reported Biot-Willis coefficientα value of 65GPa, the back-calculated Biot-Willis coeffi cient 20 0 0 1020304050 profi le decreases from 0.6 to 0.53, signifi cantly lower than Pressure, MPa the commonly assumed value of 1 while signifi cantly higher than the estimated value of 0.34 obtained from Figure 1. Grain bulk modulus and Biot-Willis coeffi cient of a sandstone sample as a func- tion of pressure calculated from its dry and brine saturated moduli [10] using Gassmann Han and Batzle’s 2004 correlation. equation. 3.3. Eff ects of input data errors on calculated grain bulk modulus 40 Measured values always have some associated errors. Velocities, especially shear wave velocities may carry 30 signifi cant uncertainties. We would like to determine the
eff ects of uncertainties from porosity, Kdry, Ksat, and Kf to the uncertainty of the predicted K . Since the relationship 20 m in Equation (7) is not linear, a Monte Carlo (stochastic)
Modulus, GPa K vacuum dry - measured simulation was used. 10 K water saturated K benzene saturated G (all saturation cases) Table 1 summarises the input parameter values K dry Gassmann [18] and their estimated ranges of uncertainties. The 0 rock sample is a Berea sandstone sample with Voigt- 0 1020304050 Differential pressure, MPa Reuss-Hill average grain bulk modulus of 39.6GPa from its mineralogical composition. All parameters were
Figure 2. Bulk and shear moduli as a function of diff erential pressure for Bedford lime- assumed to have a normal distribution with means being stone [6]. The dashed line is the dry bulk modulus calculated from Gassmann equation, the measured values and the errors represent the 95% consistently higher than the vacuum dry measured data. confi dence interval. Thus, the relative error (uncertainty) of each parameter is defi ned as: 2s 80 1.0 % error = × 100 % (17) mean
70 0.8 α Where s is the standard deviation of the parameter’s 60 sample. 0.6 50 For each set of perturbed errors, 10,000 sets of 0.4 (porosity, dry bulk modulus, wet bulk modulus, and fl uid 40 modulus) values were generated to compute 10,000 grain Biot-Willis coefficient coefficient Biot-Willis
Grain bulk modulus, GPa 0.2 30 Grain bulk modulus Km bulk moduli, which are then analysed for the mean value α Biot-Willis coefficient and standard deviation. 20 0 0 1020304050 In our base case, porosity is assigned a 1% error, Pressure, MPa Kdry and Ksat are each assigned a 3% error, and Kf carries a 10% uncertainty. The resulting K is also a Gaussian Figure 3. Calculated grain matrix bulk modulus and Biot-Willis coeffi cient of Bedford m limestone sample as a function of pressure using Gassmann equation from its water - and distribution with a mean of 44.6GPa and a standard benzene - saturated moduli [16]. deviation of 3.45GPa. The 95% confi dence interval is,
26 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Table 1. Mean (measured) values of a Berea sandstone sample [18] and ranges of uncertainties used in Monte Carlo simulations Parameters Mean values (measured) % error Standard deviation Porosity 17.6% ±1 - 15% ±0.09 - 1.3% Effective dry bulk modulus 16.8GPa ±1 - 15% ±0.25 - 1.26GPa Effective wet bulk modulus 21.1GPa ±1 - 15% ±0.32 - 1.58GPa Fluid bulk modulus (water) 2.2GPa ±0 - 30% ±0 - 0.33GPa
stochastic simulation was also performed to examine
K K effect f the eff ect of uncertainty and/or measurement errors on 60 sat φ K dry calculated grain matrix bulk modulus. The results showed K sat m
K base case that the calculated grain matrix bulk modulus is relatively constant with applied diff erential pressure (up to 50MPa) 40 Kdry effect for sedimentary rocks. However, the estimation is very K effect f sensitive to the uncertainty of dry and saturated bulk
bulk modulusbulk φ 20 effect modulus values. Our new forms of Gassmann equation can be used to eff ectively quantify the uncertainty of base case: 1% φ error, 3% Kdry & K errors, 10% K error % error% in calculated grain matrix sat f dry and saturated bulk modulus (and subsequently, the 0 0 5 10 15 20 25 30 35 seismic velocities) in fl uid identifi cation, fl uid substitution, % error of input parameters (K f /φ /Kdry/K sat) or reservoir monitoring applications. The uncertainty of the computed grain matrix bulk modulus K using Gassmann Figure 4. m NOMENCLATURE equation as functions of percent error in one input parameter (Kf, α, Kdry, or Ksat), while the remaining input parameters carry the same errors as the base case. Errors from Ksat and K: bulk modulus (GPa or psi)
Kdry have the largest impacts on the uncertainty of calculated Km. Porosity and fl uid bulk modulus, on the other hand, show negligible eff ects. Ksat: saturated bulk modulus (GPa or psi)
Kdry: dry (frame) bulk modulus (GPa or psi) therefore, from 37.7GPa to 51.5GPa (or 16% error). The K : grain (matrix) bulk modulus (GPa or psi) Biot-Willis coeffi cient α is also a Gaussian distribution with m K : fl uid bulk modulus (GPa or psi) a mean of 0.62 and a standard deviation of 0.03. The 95% f confi dence interval is from 0.56 to 0.68 (or 10% error). G: shear modulus (GPa or psi) Figure 4 shows the uncertainty of the computed grain Gsat: saturated shear modulus (GPa or psi) matrix bulk modulus Km as functions of percent error in Gdry: dry (frame) shear modulus (GPa or psi) one input parameter (Kf, φ, Kdry, or Ksat), while the remaining input parameters carry the same uncertainties as of the : Biot-Willis coeffi cient (dimensionless) base case. Errors from Ksat and Kdry have the largest eff ects φ: porosity (dimensionless) on the uncertainty of K . Minus errors in K and K (even m dry sat : density (g/cc) within laboratory measurement standard) can result in V : compressional wave velocity (km/s) a large error in the estimated value of Km. Porosity and p fl uid bulk modulus, on the other hand, show negligible Vs: shear wave velocity (km/s) infl uence. This result is not surprising, as K , φ and K f m APPENDIX A: Derivation of Equation (7) should be uncorrelated parameters. From Equation (3) we can write: 4. Conclusions
K f K f Three equivalent forms of Gassmann equation were = (1 − α ) (A.1) Km Kdry presented that can be useful for the determination of Biot-Willis coeffi cient, dry bulk modulus, and/or grain Rewriting Equation (1) as a function of gives: matrix bulk modulus of a rock. We demonstrated the ⎡ K ⎤ applicability of these equations using several sets of f 2 (K=sat − Kdry)⎢φ + (1−α ) (α −φ)⎥ K f α (A.2) published laboratory measurements and the implications ⎣⎢ K dry ⎦⎥ of the results for other estimations of rock properties. A
PETROVIETNAM - JOURNAL VOL 10/2017 27 PETROLEUM EXPLORATION & PRODUCTION
⎡ K ⎤ K equation and rearrange Equation (1) as a function of , α 2 K + f (K − K ) − (φ+1) f (K − K )α ⎢ f sat dry ⎥ sat dry the Biot-Willis coeffi cient only: ⎣⎢ Kdry ⎦⎥ Kdry ⎡ K ⎤ (A.3) f 2 ⎛ K ⎞ []K=sat − (1−α )K m ⎢φ + (α −φ)⎥ Kf α (C.1) +φ(K − K )⎜ f −1⎟ = 0 K sat dry ⎜ ⎟ ⎣ m ⎦ ⎝ Kdry ⎠ Expanding LHS and subtracting K 2 both sides, we ⎡ ⎤ f Kf Ksat 2 K f α − ⎢(φ+1) (Ksat − Kdry)⎥α have: Kdry ⎣⎢ Kdry ⎦⎥ ⎡K K ⎤ ⎡ K K ⎤ (A.4) α sat f +K φ − (=φ+1)K +φ ()K −K + K − f sat 0 ⎛ ⎞ ⎢ m f⎥ ⎢ sat m f ⎥ K f K K +φ(K − K )⎜ −1⎟ = 0 ⎣ m ⎦ ⎣ m ⎦(C.2) sat dry ⎜ ⎟ ⎝ Kdry ⎠ or equivalently, K Multiplying both sides with dry gives ⎡ ⎛ K ⎞⎤ ⎡ ⎛ K ⎞⎤ ⎜ sat ⎟ ⎜ f ⎟ K f Ksat ⎢φ()K1m−Kf − Kf ⎜1− ⎟⎥α = φ⎢()Km− Kf − Ksat⎜ − ⎟⎥ ⎣ ⎝ K m ⎠⎦ ⎣ ⎝ K m ⎠⎦ ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ α 2 − (φ+1)⎜1− dry ⎟α +φ ⎜1− dry ⎟⎜1− dry ⎟ = 0, ⎜ ⎟ ⎜ ⎟⎜ ⎟ which is Equation (15). ⎝ K sat ⎠ ⎝ K sat ⎠⎝ K f ⎠ References which is Equation (7). APPENDIX B: Derivation of Equation (13) 1. Fritz Gassmann. Uber die Elastizität Poröser Medien. Vierteljahrschrift der Naturforschenden Gesellschaftin If the same rock is subjected to two diff erent Zürich. 1951; 96: p. 1 - 23. saturation fl uids, then we have two equations in the form of Equation (7): 2. Tad M.Smith, Carl H.Sondergeld, Chandra S.Rai. ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ Gassmann fl uid substitutions: A tutorial. Geophysics. 2003; α 2 − (φ+1)⎜1− dry ⎟α +φ⎜1− dry ⎟⎜1− dry ⎟ = 0 ⎜ ⎟ ⎜ ⎟⎜ ⎟ (B.1) 68(2): p. 430 - 440. ⎝ Ksat1⎠ ⎝ Ksat1⎠⎝ K f1 ⎠ 3. Ludmila Adam, Michael Batzle, Ivar Brevik. ⎛ K ⎞ ⎛ K ⎞⎛ K ⎞ α 2 − (φ+1)⎜1− dry ⎟α +φ⎜1− dry ⎟⎜1− dry ⎟ = 0 Gassmann fl uid substitution and shear modulus variability in ⎜ ⎟ ⎜ ⎟⎜ ⎟ (B.2) ⎝ Ksat2⎠ ⎝ Ksat2⎠⎝ K f 2 ⎠ carbonates at laboratory seismic and ultrasonic frequencies. Geophysics. 2006; 71(6): p. F173 - F183. Subtracting Equation (B.2) from (B.1) gives: 4. Maurice Anthony Biot, David G.Willis. The elastic ⎛ 1 1 ⎞ ⎡⎛ 1 1 ⎞ ()φ +1 ⎜ − ⎟K α =φK ⎢⎜ − ⎟ coeffi cients of the theory of consolidation. Journal of ⎜ K K ⎟ dry dry ⎜ K K ⎟ ⎝ sat1 sat2⎠ ⎣⎢⎝ f1 f2 ⎠ Applied Mechanics. 1957; 24: p. 594 - 601. (B.3) ⎛ 1 1 ⎞ ⎛ 1 1 ⎞⎤ 5. James G.Berryman. Origin of Gassmann’s equations. + ⎜ − ⎟ − K ⎜ − ⎟⎥ ⎜ K K ⎟ dry⎜ K K K K ⎟ ⎝ sat1 sat2⎠ ⎝ sat1 f1 sat2 f 2 ⎠⎦⎥ Geophysics. 1999; 64(5): p. 1627 - 1629. 6. Gary Mavko, Tapan Mukerji, Jack Dvorkin. The rock Canceling Kdry both sides and rearranging Equation (B.3) leads to: physics handbook: Tools for seismic analysis in porous media. Cambridge University Press, Cambridge. 1998. ⎛ 1 1 ⎞ ⎛ 1 1 ⎞ φ⎜ − ⎟K =φ ⎜ − ⎟ − ⎜ K K K K ⎟ dry ⎜ K K ⎟ 7. Robert W.Zimmerman. Compressibility of ⎝ sat1 f 1 sat2 f 2⎠ ⎝ f 1 f2 ⎠ sandstones. Elsevier Science. 1991. ⎛ 1 1 ⎞ []α ()φ +1 −φ ⎜ − ⎟ 8. Luther White, John Castagna. Stochastic fl uid ⎝ Ksat1 Ksat2 ⎠ modulus inversion. Geophysics. 2002; 67(6): p. 1835 - 1843. which is Equation (13). Equation (14) then can be 9. Fredy A.V.Artola, Vladimir Alvarado. Sensitivity readily obtained by multiplying both sides by (K × K ). sat1 sat2 analysis of Gassmann's fl uid substitution equations: Some APPENDIX C: Derivation of Equation (15) implications in feasibility studies of time-lapse seismic reservoir monitoring. Journal of Applied Geophysics. 2006; If K value can be obtained (e.g. using mixture theory), m 59(1): p. 47 - 62. then one can substitute K)dry = (1 − α Km into Gassmann
28 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
10. De-Hua Han, Michael L.Batzle. Gassmann’s 15. I.Fatt. The Biot-Willis elastic coeffi cients for a equation and fl uid-saturation eff ects on seismic velocities. sandstone. Journal of Applied Mechanics. 1959; 26: p. 296 Geophysics. 2004; 69(2): p. 398 - 405. - 297. 11. James G.Berryman, Graeme W.Milton. Exact 16. Karl B.Coyner. Eff ects of stress, pore pressure, and results for generalized Gassmann’s equations in composite pore fl uids on bulk strain, velocity, and permeability in rocks. porous media with two constituents. Geophysics. 1991; 56: Ph.D. dissertation, Massachusetts Institute of Technology, p. 1950 - 1960. Cambridge, Massachusetts. 1984. 12. Keith W.Katahara. Clay mineral elastic properties. 17. Virginia A.Clark, Bernhard R.Tittmann, SEG Technical Program Expanded Abstracts. 1996: p. 1691 Terry W.Spencer. Eff ect of volatiles on attenuation - 1694. (Q-1) and velocity in sedimentary rocks. Journal of Geophysical Research. 1980; 85(B10): 13. Zhijing Jee Wang, Hui Wang, Michael E.Cates. p. 5190 - 5198. Elastic properties of solid clays. SEG Technical Program Expanded Abstracts. 1998: p. 1045 - 1048. 18. Tran Trung Dung, Chandra S.Rai, Carl H.Sondergeld. Changes in crack aspect-ratio concentration 14. Xianhuai Zhu, George A.McMechan. Direct from heat treatment: A comparison between velocity estimation of the bulk modulus of the frame in fl uid saturated inversion and experimental data. Geophysics. 2008; 73(4): elastic medium by Biot theory. SEG Technical Program p. E123 - E132. Expanded Abstracts. 1990: p. 787 - 790.
PETROVIETNAM - JOURNAL VOL 10/2017 29 PETROLEUM EXPLORATION & PRODUCTION
MERGING OF 3D SEISMIC DATA Ta Quang Minh, Nguyen Thanh Tung, Mai Thi Lua, Bui Thi Hanh, Do The Nam, Nguyen Ngoc Huy Vietnam Petroleum Institute Email: [email protected]
Summary
Merging of individual seismic datasets into a single uniform dataset is useful for structure analysis, geological modelling, and seismic attribute analysis for a large area. However, many challenges are met, both from the acquisition aspect to processing of data before merging. The authors investigate in detail the technical diffi culties of merging data, and discuss methods to accomplish the goal with an emphasis on post stack seismic merging, as well as merging results from our processing of fi eld seismic datasets. Key words: Merging seismic data, seismic data processing, special seismic techniques.
1. Introduction structures’ orientation. A more feasible alternative is to merge available “small” overlapping 3D vintage seismic data into a single Seismic survey data has the pioneering role dataset with uniform amplitude, frequency, and phase. A merged in laterally extending the geological study area. seismic dataset should provide a reasonable approximation to A regional-scale seismic study, such as the study full area acquired/processed data at a much economical cost and of a basin, may face a stiff challenge where the shorter operation/processing time. area is so vast that it might not be coverable by a single seismic survey. In order to do so, the In fact, seismic data merging has been performed regularly ability to combine and blend information from on many foreign projects [1 - 3] and the results have been very individual surveys becomes necessary. encouraging. Figure 1 shows an example map of several overlapping surveys in India [1] subjected to the merging process. Nowadays, a number of 3D seismic surveys has been carried out each year on the There are several important applications of merged seismic continental shelf of Vietnam with a typical data. The subsurface image of regions near the edges of a local coverage area ranging from about 100km2 to survey is usually distorted from the migration due to the defi ciency 3,500km2, usually within a study block. The of input seismic data, which eventually can be compensated by division of the continental shelf into blocks and another overlapped survey data [1]. The most important application assignment to diff erent operation companies is probably the ability to carry out the interpretation of structures also implies that each 3D seismic survey is continuously and seamlessly on the merged seismic data from one limited to the block operated by a company. A local region to another, and hence enable many geological analyses regional-scale geological assessment, therefore, for a large area such as the structural analysis, sequence stratigraphy will be very diffi cult since the data is a collection analysis, etc. Moreover, many attribute extraction applications can of patchy seismic surveys that often have also be performed on the merged dataset such as various seismic diff erent acquisition characteristics and thus attribute analyses, PaleoScan software, etc. distinctive recorded seismic data characteristics. Certainly, reacquiring seismic data for the whole region and processing the data as a large single seismic dataset may solve part of the problem of patchy data. However, this is unrealistic and prohibitive due to expensive costs and long computation time, not to mention the regional administrative problems (a) (b) as well as diffi culties with the survey design Figure 1. (a) Display all fi ve surveys acquisition grids (b) Display data acquisition superimposed on phase to comprehensively cover various local Bathymetry [1]
Date of receipt: 7/9/2017. Date of review and editing: 7/9 - 21/9/2017. Date of approval: 4/10/2017.
30 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
2. Technical diffi culties with merging seismic data The most obvious diff erences are due to the acquisition confi guration on the survey boat such as The process of merging seismic data is basically related to the source confi guration, including the types of the the compensation for the diff erences in seismic characteristics source - air gun, vibroseis, dynamite, the geometry of of various overlapping datasets. The variation in seismic the source array, and the volumes of individual guns, characteristics between datasets can broadly be categorised etc. In addition, the variation in the confi guration of as coming from 2 major sources: the widely varying seismic the receiver side may include the number and length acquisition confi gurations and the diff erences in processing of the streamer cables, the interval of the receivers, technologies. We will discuss the eff ect of each type of the type of receivers such as hydrophones, velocity variations on the seismic characteristics as follow. phones, the recording and sampling intervals, the 2.1. Eff ect of variations in acquisition schemes frequency of anti-aliasing fi lter, etc. Those diff erences provide decisive impacts on the amplitude and Each seismic survey is usually designed with a specifi c goal frequency content of the acquired seismic data, the suitable for a particular local region. Thus, diff erent seismic signal to noise ratio (SNR), as well as the fold density surveys in nearby regions may have diff erent acquisition at local common middle points. Drastic amplitude parameters. We concentrate on a few important acquisition diff erences between two overlapping surveys can be parameters. seen in Figure 2. The frequency content diff erences may come from another source. Nearby surveys may have diff erent source depths and receiver depths that can aff ect the ghost frequencies. Typically, the ghost notch frequency is related to the source or receiver depth by the formula [4]
≈ water notch 2
Where Vwater is the seismic velocity in the water medium, d is the depth of the source or receiver
and fnotch is the notching frequency. The eff ect of the ghost notch on the frequency content is illustrated Figure 2. Drastic amplitude diff erences between two overlapping surveys in Figure 3. The frequency limitation by the ghost notches, Bubble pulse In- Constructive primary-ghost in turn, will aff ect the signal frequency bandwidth, terference interference which will impact the seismic resolution as well as the wavelet shape. Seismic frequency content and resolution diff erences, as a combination of several factors, can be seen in Figure 4 where the survey on the left has a much lower frequency content and resolution, comparing with the survey on the right. New acquisition advancements such as PGS’s GeoStreamer/GeoSource [5, 6] or CGG-Veritas’
Amplitude BroadSeis [7] technologies will further widen the Destructive primary-ghost interference seismic frequency bandwidth/resolution diff erences 0 100 200 300 comparing to the seismic data acquired from the Frequency (Hz) traditional hydrophone - fl at cable confi guration.
Figure 3. The impact of the ghost notching frequency on the frequency content of the acquired The presence of external factors such signal [4] as acquisition time (tidal change), changing
PETROVIETNAM - JOURNAL VOL 10/2017 31 PETROLEUM EXPLORATION & PRODUCTION weather (rough vs quiet sea for marine acquisition), changing A parameter that is also frequently changing unconsolidated layer thickness (for land acquisition) might result from survey to survey is the acquisition azimuth/ in another class of problems which is the diff erences in phases and orientation. This leads to the diff erences to the statics between overlapping surveys. The phase/static diff erences orientation of the migration grid (inline-crossline are troublesome for interpretation horizons to align from a seismic grid) in the post-acquisition seismic processing dataset to another. The eff ect and correction for phase/static workfl ow. A typical 3 survey overlapping grid problem is discussed further in Section 3. confi guration is shown in Figure 5. As we can see, each grid system has a diff erent inline orientation, as well as diff erent grid binning dimensions, not to mention the possible CMP- fold diff erences. The discrepancy in the grid orientations eventually leads to the need to re-grid seismic datasets in to a common encompassing grid system (master grid) and re-number the inline and crossline during the seismic merging process (Section 3). A less obvious eff ect of the diff erences in the sail-line’s direction is related to the amount Figure 4. The diff erence of the signal frequency bandwidth/resolution of seismic illumination. For a single survey orientation, underground geological objects may have certain seismic features being illuminated better than other depending on the alignment of the object and the sail-line’s direction. This can be illustrated in Figure 6 where, as part of the survey design process at the VPI, we simulate the seismic illumination of a target horizon/ fault under diff erent sail-line azimuths of the survey boat with other acquisition parameters kept the same. As we can see a major fault is illuminated diff erently under diff erent sail-line orientation. The diff erences in the amount of seismic illuminations on the same geological object mean that the images of the same object on the fi nal overlapping seismic cube may have diff erent apparent, amplitude and image quality, which requires the amplitude conditioning before merging. Figure 5. Location map showing the Grid orientation and Master grid
(a) (b) (c) Figure 6. Diff erent seismic illumination on the same fault depending on the sail line azimuth: a. 168 degree, b. 258 degree, and c. 319 degree. Illumination by the 258 degree sail line azimuth appears to create the clearest observation of the fault surface.
32 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
2.2. Diff erences in seismic data processing routines
The diff erences in processing workfl ows from various contractors should also present some problems with merging seismic surveys. Each processing jobs may have dramatically diff erent objectives, and technologies including:
- Diff erent target objects that may lie at diff erent depth (shallow vs. deep).
- Diff erent technologies used in the process, such as PSTM, PSDM, and CBM, etc. Figure 7. Seismic event misalignments of two seismic survey - Diff erent parameters used in the fl ows (possibly on the same dataset). We highlight some major aspects, including phase/static, amplitudes, frequencies, and SNR, that the diff erences impact the merging process. A common issue is related to the misalignment of events between two overlapping datasets, also known as the static/phase problem. Beside those Figure 8. Technology diff erences comparison - FX Migration vs PSTM Migration originated from the survey confi guration diff erences, the causes of the misalignment might also be traced to several processing routines:
- The diff erence in velocities used in the seismic migration might lead to some events (dipping refl ection planes) to be migrated to a diff erent time/depth comparing to one processed from the other routine. Figure 9. Technology diff erences comparison - PSDM vs CBM
Figure 10. Combination of the diff erences in seismic acquisition and processing may result in two overlapping seismic sections (left and ight)r with completely diff erent appearances
PETROVIETNAM - JOURNAL VOL 10/2017 33 PETROLEUM EXPLORATION & PRODUCTION
- The misalignment may also come from the up comparing to the vintage. The example shown in Figure 8 diff erences between the time and depth migration, is the comparison of the seismic quality from results processed and the dipping features may not align for the PSTM by the old FX migration in 2001 (left) and reprocessed by the and PSDM converted to time. PSTM algorithm in 2002, in which, the quality of PSTM seismic data is signifi cantly improved, in terms of details (resolution) - Another source of misalignments is from the and frequency bandwidth, compared to the old (possibly diff erent level of Q-compensation between datasets obsoleted) FX migration seismic data. - leading to diff erent phase shifts. Figure 9 illustrates another comparison between the old - Events at the edges of a survey is usually PSDM and the newer CBM processing, where many features, observed to have incorrect imaging time/depth such as faults, top basement show up more clearly on the fi nal due to the insuffi cient input seismic points for the CBM stack section, comparing to the vintage PSDM processing. migration. Overlapping seismic datasets with diff erent level of SNR will Seismic event misalignments are serious because cause diffi culties in matching up data, or lining-up events. A in case you are willing to align those events at a certain possible remedy is to perform seismic enhancement for poor depth, events at other depths may still misalign. quality data such as one described in a recent study from our Another discrepancy is about the amplitude group [8], although the results might be limited after stacking has been performed. Or better, input seismic data might be between overlapping seismic data. Beside the inspected before merging to select those with similar SNR. cause from the acquisition confi guration factors (as discussed above), the variation in the seismic The accumulation eff ect of various types of diff erences amplitude might be the consequence of the from the survey confi guration to seismic signal processing diff erence in migration apertures (each image point may result in seismic sections of the overlapping region with receives contributions from a varying number of signifi cant diff erent appearances such as one in Figure 10. In traces). The processing fl ow may further contribute this case, a decision will need to be made regarding the merge to it by various amplitude compensation such as border, which, in turn, will decide which part of the individual the post stack time-variant scaling, Automatic Gain survey will remain in the merged dataset, and which part will Control. Amplitude compensation for the absorption be thrown away. (Q-compensation) might be diff erent in workfl ows 3. Merging 3D post-stack data from diff erent processing contractors. The diff erences in the amount of Q-compensation The goal of the merging of 3D post-stack seismic surveys might also be the source of diff erences in the is to create a single 3D seismic cube which balances the phase frequency content of two overlapping surveys. and amplitude of individual stacked input seismic cubes. For Besides, frequency manipulation techniques such geological structure and seismic attribute analysis, the results as spectral shaping, spectral whitening, spectral might be adequate, but certainly cannot be as good as merging bluing, source deconvolution, and source wavelet pre-migration seismic data. The workfl ow can be summarised de-ghosting, etc. might aff ect the frequency content in 5 major processing steps as shown in Figure 11. Note that between two datasets. However, for the marine data, the fl ow chart is over-simplifi ed and each major step can be these drastic frequency-altering techniques are rerun as needed. not usually off ered by the processing contractor to 3.1. Regrid of seismic data preserve the original frequency bandwidth (unless requested by the client). As discussed in Section 2, due to the diff erences in survey orientations, overlapping surveys may have diff erent migration An often-overlooked aspect is the merging of grids with diff erent azimuths and diff erent CDP spacings. Re- datasets with diff erent SNR, which might make the fi nal results look synthetic (a poor dataset is stitched to a good dataset). Newer computing hardware resources enable more advanced processing/ imaging algorithms that can provide seismic results with enhanced SNR and new seismic features shown Figure 11. Five major steps fl owchart in merging post-stack seismic data
34 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
gridding of seismic data is essentially envelopes of one dataset to match with that of the other one. Since there spatial resampling (i.e. interpolation) could be diff erent gain functions applied to the datasets, the amplitude of data by points on a common correction is usually time-variant too. Figure 13 illustrates the results encompassing master grid (Figure 5). from amplitude conditioning where the amplitude of the left section is In our experiences, the selection of the matched to the one on the right section. Care has been taken to preserve master grid may fall into one of the two relative amplitude so that seismic anomalies (such as Direct Hydro Carbon popular choices: Indicator (DHI)) still show up correctly after amplitude conditioning, as the
- We can select the orientation of the majority of the surveys, which has the benefi t of having a smaller number of surveys to be regridded.
- Or we can select the orientation of the master grid to be orthogonal to the major fault or features of concern. Various interpolation schemes exist such as bilinear interpolation, bicubic interpolation, polynomial interpolation or spline interpolation, etc. We chose the sinc function interpolation (Whittaker- Shannon interpolation) since it is the perfect reconstruction of the original (continuous) signal according to the Shannon sampling theorem [9]. In one dimension, the interpolation takes the form of:
Figure 12. Basemaps grid point before and after re-gridding Where t is the continuous time/ space variable, sinc(t) = sin(πt)/(πt) and T is usually the sampling rate that gives the samples {x[n]}. From the interpolation formula, resampling simply means to choose to resample t at a diff erent discrete rate: t = T1, 2T1,…, nT1,… In our case of spatial resampling, we use the Figure 13. Two merging seismic sections before and after amplitude conditioning extended 2D version of the formula. An illustration of re-gridding can be seen in Figure 12 where spatial data points that misalign with the master grid are resampled into points on the master grid (located at the centre of each new cell).
3.2. Amplitude conditioning
Amplitude equalisation between (a) (b) Figure 14. A stack section with a bright spot before (a) and after (b) amplitude conditioning. Notice the range of overlapping seismic dataset usually amplitude of the original seismic is about -4e5 to 4e5, while the range of amplitude of the merging seismic is about means correcting the amplitude -4 to 4. The bright spot has a relatively high amplitude in both cases.
PETROVIETNAM - JOURNAL VOL 10/2017 35 PETROLEUM EXPLORATION & PRODUCTION
Figure 15. Average amplitude spectra from diff erent campaigns [10] amplitude conditioning step only applies a constant smooth gain function (for the most part of the merging seismic cubes). As illustrated in Figure 14, a bright spot remains relatively standing out after amplitude conditioning.
3.3. Frequency conditioning Some special regions may require frequency conditioning due to the obvious diff erences in the frequency spectra. Thus, the frequency conditioning step aims to minimise the diff erence in amplitude spectra (rather than the energy in the diff erent section). This is usually accomplished by a matching or shaping fi lter. The process of frequency matching may contain an inherent confl ict goal. On the one hand, we want uniform Figure 16. Two merging seismic sections before and after static conditioning frequency bandwidth on the merged cube for the attribute content on the deeper section. Thus, techniques such analysis on the fi nal result and the easiest way is to choose as to apply inverse Q-compensation and reapply the seismic cube with the lowest frequency bandwidth as the an equal amount of Q can also be carried out if reference and band-limit the frequency of other higher-quality information from the processing report is available. cubes. Certainly, this approach will degrade the quality of some input seismic. On the other hand, we also desire to preserve 3.4. Static/phase conditioning as much frequency bandwidth/time resolution as possible, This step is to match the timing and phase this implies to use the seismic cube with wider bandwidth as of refl ection events on the overlapping region. the reference and to perform shaping fi lter to enhance the Static/phase conditioning is crucial to merging and frequency content of the seismic data with lower bandwidth responsible for the smooth/seamless look of the (whitening the data). However, the result might be limited as merged cube. In the actual processing, the fi rst thing whitening the data too much will amplify the noise. In drastic to consider is to make sure all seismic dataset has the cases, a decision on a compromise frequency bandwidth same polarity. This is usually obvious by inspecting might be made. major events. Optionally, performing Q-conditioning An example of performing frequency conditioning can on the phase, again, might also be helpful to improve be seen in Figure 15 [10] where the authors apply spectral alignments of events on the deeper sections. whitening to smooth out frequency spectra across the datasets. Static/phase conditioning, then, try to align The diff erences in the amount of Q-compensation might be events by the sub-sample levels. Techniques to another aspect that is crucial to the discrepancy in the frequency estimate the amount of misalignment may include
36 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
the time correlation or the phase spectra matching. This conditioning step then adjusts the data by static shifts and phase rotation. In our experiences, events at the outermost edges might not be easily matched due to the migration distortion, but they are usually discarded anyway in the next step. Figure 16 shows the eff ect of matching the statics from two diff erent datasets. As we can see, major events can be matched and, hence, run continuously between the two surveys. However, the result is not perfect as some small residual artifacts of the borders remain. This is the basic limitation of performing merging on the post-stack dataset as static/phase conditioning may not be able to match all events from the shallow to the deeper intervals.
3.5. Stitching fi nal results Figure 17. Two merging seismic sections before and after merging process
The last step is to merge conditioned seismic cube together. This involves border design and throwing away redundant seismic data (as shown in Figure 10). The usual added benefi t is that discarded seismic is usually the data at the edges of the surveys where major distortions occur.
4. Field data results Figure 18. Time slice of two merging seismic sections before and after merging process We present fi eld data results from the merging process performed at EPC-VPI. Some more merging results are shown in Figure 17 where we had a seismic section before and after the merging process. As we can see, events are continuous and the whole merged section has a uniform amplitude, giving the perception of a dataset from a single seismic acquisition. The continuity of the events can be further illustrated in Figure 18 where we took a time slice before and after the merging. In this case, after merging, the border disappears and events are continuous from one side of the border to the other side. Figure 19. RMS analysis on the merged seismic cube
PETROVIETNAM - JOURNAL VOL 10/2017 37 PETROLEUM EXPLORATION & PRODUCTION
We also attempted to run an RMS seismic attribute Technically, reprocessing and merging from raw data extraction from the merged seismic cube on a horizon should provide a better result due to the combination of within the Oligocene layer and the result is successful a common processing/conditioning routine, a common (Figure 19). The attribute map looks continuous velocity analysis, a common migration grid, and a common throughout the merge borders, allowing an interpretation migration algorithm. The advantages of the premigration- of features and abnormal amplitudes, although some prestack merge results comparing to the post-stack merge results can be shown in Figure 20 [13] where minor artifacts from a very small overlapping area (low results from the premigration-prestack seismic merging fold edge region) may remain. show signifi cantly more details in the overlapping region 5. Towards 3D premigration pre-stack seismic data due to the added complementary data for migration. merging Another illustration is shown in Figure 21 [14] from Sino Geophysical with similar observations. As we have seen in Sections 3 and 4, the problem of Many issues that have been discussed in Section misalignments of seismic events can be partially solved 2 may remain with premigration-prestack data merging by matching phase/static from post-stack seismic cubes such as the discrepancies in the shot records such as shot (Section 3.2). However, residual artifacts are frequently length, shot density, or receiver intervals, cable length observed, especially on the dipping events. This problem which can be solved by the shot record regularisation. can only be solved completely by performing a merge on Another discrepancy might be related to the CDP folds the premigration pre-stack seismic data (merging before (midpoint density) and require a CDP fold regularisation imaging) such as the studies in [10 - 13]. also. Usually, trace regularisations (such as the shot
Figure 20. A comparison of post-stack and pre-stack merging [12]
Old post stack merging section New prestack merging section by WEFOX Figure 21. A comparison of post-stack and pre-stack merging [14]
38 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
record or cdp fold regularisation) can be accomplished merging workfl ows, which should promise better, more by the combination of the trace decimation and 5D trace uniform, more seamless merging results, although at a interpolation. possibly much more expensive computational cost, more complex algorithms/workfl ows and longer run time. Pre-stack merging of seismic data will be very useful for pre-stack analyses such as AVO and inversion References analyses, although at the cost of some serious technical diffi culties. Common sense indicates that the process of 1. S.Basu, S.N.Dalei, D.P.Sinha. 3D seismic data merging merging premigration data requires huge processing - A case history in Indian context. Geohorizons. 2008. resource requirements (machines and workforces), a 2. Agus Widjiastono, Nasser Hamrbtan, Almhedwi long processing time, and a high cost. Various issues like Aljilani. Merging 3D seismic data for IR fi eld development. seismic illumination and frequency bandwidth might still North Africa Technical Conference and Exhibition, Cairo, be present. Egypt. 14 - 17 February, 2010. 6. Conclusions 3. M.Radovcic, S.Cumbrek, I.Nagl, T.Ruzic. 3D seismic data merging: A case study in the Croatia - Hungary border 3D merging of seismic surveys recorded in diff erent area. 6th Congress of Balkan Geophysical Society, Budapest. acquisition processes with various acquisition geometries, 3 - 6 October, 2011. orientation and processing workfl ows is a very intricate process. Major problems are the mismatch of amplitude, 4. Mamdouh R.Gaddalah, Ray L.Fisher. Applied frequency bandwidth, static/phases and SNR between seismology: A comprehensive guide to seismic theory and overlapping seismic dataset. Fundamentally resolving application. PennWell Books. 2005. each problem is necessary to achieve the goal of 5. PGS. Geostreamer. www.pgs.com. treatment and help the interpreters to meet the geological objectives. Our hunt for the solutions has resulted in an 6. PGS. Geosource. www.pgs.com. 2017. in-house merging workfl ow for the post-stack data that is 7. CGG. Broadseis. www.cgg.com. 2017. relatively fast to execute and cheap to implement on the 8. Tạ Quang Minh, Bùi Thị Hạnh, Nguyễn Tiến Thịnh. available computing hardware. Results processed at the Phân tích cấu trúc và nâng cao chất lượng tài liệu địa chấn. Vietnam Petroleum Institute (VPI) show a suffi ciently good Tạp chí Dầu khí. 2017; 8: trang 16 - 24. merged dataset that meets the requirements of uniform amplitudes/frequencies/phases, with a seamless look of 9. Alan V.Oppenheim, Ronald W.Schafer, John R.Buck. the sections/time slices, and refl ection events running Discrete time signal processing (2nd edition). Prentice Hall. continuously from one area to another. Our resultant 1999. merged datasets thus look quite promising that they 10. Bera. Prestack 3D land data merging a case history could be used for seamless geological interpretation or from south Asian shelf India. SPG India - 11th Biennial continuous seismic attribute analysis. The limitations of International Conference and Exposition. 2015. the post-stack merging process including some dipping events might not be to align very well, some artifacts may 11. Wang Lixin. The pre-stack merging imaging remain due to small overlapping area (low fold at the edge/ techniques and its applications. SEG Technical Program migration distortion), and sometimes high-quality data Expanded Abstract. 2010: p. 2870 - 2874. may need to be downgraded to match low-quality data. 12. Jun Cai, Manhong Guo, Shuqian Dong. Merging Those imperfections, although looking minor, sometimes multiple surveys before imaging. SEG San Antonio 2011 may have serious consequences in applications such as Annual Meeting. 2011. ambiguity in interpretation or producing seismic anomaly 13. A.K.Srivastav, C.Chakravorty, K.V. Krishnan. Some in attribute analysis, thus extreme care must be taken issues in 3D Pre-stack merging. 8th Biennial International both during the process of performing post-stack seismic Conference & Exposition on Petroleum Geophysics. 2010. merging and during QC and recommendations, as well as during actual usage. 14. Sino Geophysicals Co. 3D pre-stack merging processing technology. www.sinogeo.com. 19/9/2017. Looking towards the future, we would like to explore the implementation of a 3D prestack-premigration
PETROVIETNAM - JOURNAL VOL 10/2017 39 PETROLEUM EXPLORATION & PRODUCTION
QUICK PRE-STACK SEISMIC INVERSION TO PREDICT RESERVOIR PROPERTIES AT A GAS AND CONDENSATE FIELD IN NAM CON SON BASIN Phan Dang Thai Son1, Phan Tien Vien2, Tran Hong Nam1, Ngo Anh Quan1, Hoang Minh Hai1 1Bien Dong POC 2Vietnam Oil and Gas Group Email: [email protected]
Summary
This paper introduces a new approach to quickly predict reservoir quality along a particular directional well-path, to justify the optimisation process for well location. The workfl ow employs a simulated annealing algorithm to convert available seismic data at well locations into some petrophysical properties that indicate reservoir quality such as porosity and water saturation. The inversion process is constrained by several seismic traces that are previously extracted along the well-path, which help signifi cantly reduce the calculation cost. The results were successfully applied in two development wells drilled into gas bearing reservoirs in Nam Con Son basin, off shore Vietnam.
Key words: Reservoir quality, seismic inversion, rock physics model, Vshale, porosity, water saturation, Nam Con Son basin.
1. Introduction
Seismic inversion (post- and pre-stack) is widely used in reservoir characterisation [1 - 4]. The basic idea is to relate changes in rock properties due to hydrocarbon introduction into the formation to the changes in pre-stack seismic amplitude via a mathematical equation, which is also called a rock physics model. The model is then used to convert rock properties (porosity, saturation, shale content) into some common seismic properties (velocity, density, Poisson ratio) [4], which are further used to predict other properties via cross-plot or some mathematical approaches such as neural network or multi-attribute cross-plots [5 - 7]. A gas and condensate fi eld in Nam Con Son basin is recently developed to produce hydrocarbon from Miocene reservoirs. Geophysical studies show strong AVO responses at reservoir levels, which indicate hydrocarbon saturated sands. The fl uid replacement modelling using Gassmann fl uid substitution equation showed signifi cant reduction of off set amplitude when Figure 1. Fluid replacement modelling shows improved AVO anomalies of gas replacing water gas replaces water within the reservoir bodies reservoirs, including AVO classes, fl uid factors indicator, and the AB (Figure 1). fl uid factor (ABFF) to help improve the interpretation and well target Intense in-house qualitative studies [8] optimisation. In Figure 2, the ABFF attribute successfully highlights have been carried out to understand the the fl uid existence in reservoirs A1 and F1.
Date of receipt: 16/2/2017. Date of review and editing: 16/2 - 23/10/2017. Date of approval: 24/10/2017.
40 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Seismic inversion is practically employed to quantitatively predict reservoir properties at candidate well locations. The common approach is performed on the trace by trace basis, which is suitable for nearly vertical well-path as a single run could provide suffi cient information to the whole well-path. However, since most wells in the fi eld are deviated, it requires many seismic traces to cover the well profi les, consequently increasing Figure 2. ABFF attribute highlights fl uid potential of reservoirs A1 and F1 the calculation cost. A new approach is proposed to Table 1. Input petrophysical parameters for the rock physics model quickly estimate the interested properties and help Parameters Sandstone Shale Water Hydrocarbon save the decision making time.
Vp (m/s) 4,300 3,350 1,700 700 2. Method
Vp/Vs 1.55 1.85 Density (g/cc) 2.67 2.6 1.01 0.00083 The approach employs Wyllie’s linear approximation equation [9] as a rock physics model to connect the petro-physical properties - porosity
(PHI), water saturation (Sw) and the shale content
(Vsh) - to geophysical parameters such as velocity
(P-wave Vp and S-wave Vs) and density. The latter set of parameters is then used to convert into synthetic responses using the re-written three-term Aki- Richard approximation by Wiggins et al. [10]. Below is a set of Wyllie approximation equation used in our method:
Figure 3. Excellent correlations in QC plot of the input rock physics model
Where: Input data:
Off set gathers + well logs Vpst, VpSh, Vpw, VpHC are P-wave velocities of pure Pre-stackdata (V , P , S ) sh or w sandstone framework, pure shale, water and Property model cubes hydrocarbon; Intercept [A] + gradient [B] (Z , Z , density) using wiggin's + full-stack cube + wavelets p s equation (Inter/Extrapolation) Dnst, DnSh, Dnw, DnHC are densities of pure sandstone framework, pure shale, water and Extract data along well bore hydrocarbon. The parameters used at this fi eld are listed as Iterative Inversion in Table 1. Figure 3 shows the comparison between
the estimated Vp, Vs and density curves (red line) with the actual recorded log (green line). Excellent Output curves: V , P , S sh or w matches between corresponding curves validate the above rock physics model. Figure 4. Inversion fl ow chart
PETROVIETNAM - JOURNAL VOL 10/2017 41 PETROLEUM EXPLORATION & PRODUCTION
The simulated annealing algorithm [3, 11] is used to Where: invert the pre-stack seismic data. This global optimisation process takes into account a low frequency model and perturbs it to fi nd the best fi t model that satisfi es the seismic responses. The outputs are a best fi t model and many other fi t models that can be used for uncertainty analyses. In this study, we use the best fi t model for fi nal prediction values. Figure 4 summarises the general workfl ow of the inversion algorithm. 3. Results The low frequency starting models are generated by interpolating logs from available wells nearby, and We successfully applied the method to a drilled applying a low frequency fi lter to retain the low frequency nearly vertical exploration well (well-X) for quality control content. Figure 4 demonstrates the inversion fl ow chart and on two other development locations (wells P1 and template. P2) with very close results from fi nal LWD petrophysical analyses. The inversion is expected to output three parameters. In order to remove the non-uniqueness in the solution, the The nearly vertical exploration well (well-X) indicated input seismic data has to be of multiple traces. Therefore, gas bearing for two target reservoirs: a gas saturated sand pre-stack data is required. A1 of 15mTVD thickness with 26% porosity, 20% average
water saturation and 5% shale volume (Vsh); a massive gas The common pre-stack inversion involves comparing and condensate saturated sand F1 of 36mTVD thickness synthetic to real angle gather traces to update the input with 14% porosity, and 46.9% average water saturation. to satisfy the real data [4, 12]. However, this is very time- In Figure 5, the inverted results (dark blue) have excellent consuming considering many angle traces involved in the matches in stack, intercept and gradient traces, while calculation process at each location. For this approach, the porosity, S and V , fi t very well with actual curves to reduce the computing cost, we convert the NMO w shale calculated from well logs (red). The estimated Vshale and corrected off set volume into intercept, gradient and porosity curves even give accurate prediction of the curvature volumes using an input stacking velocity cube layering from the changes in values. This encourages us [4]. The intercept corresponds to seismic response of a to apply the approach to predict at other deviation well zero off set trace, the gradient and curvature refl ect how locations. amplitude changes with off set. At the fi rst development location, well-P1 targeting Due to the limitation of data acquisition parameter, both A1 and F1, maximum 33o deviated profi le, the the pre-stack volume is limited to 30 degree angle off set synthetic traces match excellently with the input for usable quality. With this limited angle range, the eff ect traces (Figure 6). The inversion predicted very well the of the curvature term on the overall inversion results properties, with almost perfect match of the logs of A1; is minimal. Meanwhile, the fi nal stack volume, which excellent matches of porosity and Vshale and average match is a summation product of the intercept, gradient and of the water saturation of the F1. With the latter reservoir, curvature, is still showing a strong eff ect on the inversion the algorithm also predicted layering via fl uctuations results. Therefore, we use the full-stack volume along of porosity and Vshale curves. The actual drilling results with the intercept and gradient volumes for inversion confi rm this point. constraint. To perform the inversion at deviated wells, we fi rst extract those parameters along the well-paths and On the second development well-P2, targeting o input them into the algorithm. reservoir F1 with maximum 42 deviated profi le, the inversion predicted reservoir existence at target Three synthetic traces created by using the locations, with high correlations with actual seismic equation of Wiggins et al. [10] are compared against traces. However, the estimated properties are poorer the corresponding traces from real data to control the than the actual well results of the F1 interval (Figure 7). inversion algorithm. Below Wiggins’ equations are used to This suggested the heterogeneity of F1 properties from generate synthetic traces for error calculation. well-X location, such as changes in signatures of Vshale
and Sw (much lower readings). Since the prediction is
42 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
dependent of the input model, which is controlled by the well-X properties, the error is reasonable. 4. Conclusion
The introduced method off ers a quick-look prediction of reservoir quality along a proposed deviated well path by directly inverting the seismic data into reservoir properties. It requires a starting model acting as an initial guess, guiding the algorithm to the optimal result that satisfi es seismic traces extracted along the well-bore. The starting model can be generated from interpreted petrophysical parameters of a nearby well. This method is proven to be reliable at well Figure 5. Inverted dark blue curves having excellent matches with input seismic traces locations within homogeneous area. However, and actual logs (red curves) at well- X for highly heterogeneous reservoirs, the method will not well predict the properties unless proper guidance of the geological heterogeneity is given to the input model. At the current stage, the algorithm is working along a particular well-path. In case there are enough controlling wells, the algorithm could be used for 3D volumes for spatial mapping of reservoir properties. Acknowledgement
We thank the Vietnam Oil and Gas Group and Bien Dong POC for allowing us to use the data and publish the results. Reference Figure 6. Inverted dark blue curves with excellent matches with input seismic traces and actual logs (red curves) at well- P1 1. Albert Tarantola. Inversion of seismic refl ection data in the acoustic approximation. Geophysics. 1984; 49(8): p. 1259 - 1266. 2. Brian Russell, Dan Hampson. A comparison of post-stack seismic inversion methods. SEG Technical Program Expanded Abstracts. 1991: p. 876 - 878. 3. Mrinal K.Sen, Paul L.Stoff a. Global optimization methods in geophysical inversion. Elsevier Science Publications, Netherlands. 1995. 4. Daniel P.Hampson, Brian H.Russell, Brad Bankhead. Simultaneous inversion of pre-stack seismic data. SEG Technical Program Expanded Abstracts. 2005: p. 1633 - 1637. 5. David M.Dolberg, Jan Helgesen, Tore Hakon Figure 7. Inverted dark blue curves with having excellent matches with input seismic traces but average matches with actual logs (red curves) at well- P2 Hanssen, Ingrid Magnus, Girish Saigal, Bengt
PETROVIETNAM - JOURNAL VOL 10/2017 43 PETROLEUM EXPLORATION & PRODUCTION
K.Pedersen. Porosity prediction from seismic inversion, 9. M.R.J.Wyllie, A.R.Gregory, L.W.Gardner. Elastic wave Lavrans fi eld, Halten Terrace, Norway. The Leading Edge. velocities in heterogeneous and porous media. Geophysics. 2000: p. 392 - 399. 1956; 21(1): p. 41 - 70. 6. Joseph Christian Adam Frank Valenti. Porosity 10. Ralph Wiggins, Geogre S.Kenny, Carrol D.McClure. prediction from seismic data using multi-attribute A method for determining and displaying the shear-velocity transformations, N Sand, Auger fi eld, Gulf of Mexico. The refl ectivities of a geologic formation. European Patent Pennsylvania State University, College of Earth and Application 0113944. 1983. Mineral Sciences. 2009. 11. William L.Goff e, Gary D.Ferrier, John Rogers. Global 7. Josimar Silva, Gorka Garcia, Viviane Farroco, Elita optimization of statistical functions with simulated annealing. de Abreu, Andrea Damasceno. Joint estimation of reservoir Journal of Econometrics. 1994; 60(1, 2): p. 65 - 99. saturation and porosity from seismic inversion using 12. James L.Simmons, Milo M.Backus. Waveform- stochastic rock physics simulation and Bayesian inversion. based AVO inversion and AVO prediction-error: Geophysics. 12th International Congress of the Brazilian Geophysical 1996; 61(6): p. 1575 - 1588. Society & EXPOGEF, Rio de Janeiro, Brazil. 15 - 18 August, 2011: p. 1327 - 1330. 13. Steven R.Rutherford, Robert H.Williams. Amplitude-versus-off set variations in gas sands. Geophysics. 8. Han N.Tran, Guy Peterson, Nam H.Tran, Lam 1989; 54(6): p. 680 - 688. Q.Nguyen, Hai M.Hoang. Well location optimization using seismic AVO analysis in turbidite reservoir gas sands, Block 14. Ravi P.Srivastava, Mrinal K.Sen. Stochastic 05-2, Nam Con Son basin, off shore Vietnam. PVEP Technical inversion of pre-stack seismic data using fractal-based initial Forum “Challenging Reservoirs in Vietnam”, Ho Chi Minh models. Geophysics. 2010; 75(3): p. 47 - 59. city. 29 - 30 May, 2013.
44 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
APPLICATION OF GEOCHEMICAL TECHNIQUE TO REDUCE ALLOCATION COST FOR COMMINGLED PRODUCTION WELLS FROM MULTIPLE RESERVOIRS Ha Thu Huong1, Kieu Anh Trung1, Nguyen Minh Quy1, Hoang Long1, Le Thi Thu Huong1, Tran Van Lam2 1Vietnam Petroleum Institute 2PVEP-POC Email: [email protected]
Summary
This paper presents a new technique which has been developed to achieve production allocation at a reduced cost by the Vietnam Petroleum Institute (VPI). A case study of using geochemical fi ngerprint technology to back allocate commingled well stream from multiple reservoirs in an off shore basin, southeast Vietnam has been carried out. A large number of oil samples were collected over time and analysed by gas chromatography for their fi ngerprint and production allocation. Quantitative results were attained using a proprietary computer program which mathematically calculates the relative contributions of end member oils based on chromatographic data. The geochemical allocation results compare very favourably with the metering data. The success of the geochemical allocation allows the cost of fi eld operation to be reduced and eff ective support to be given to the production management. Key words: Back allocation, gas chromatography, geochemical fi nger-printing.
1. Introduction method to re-assess and crosscheck individual fi eld/ reservoir contribution to commingled pipeline oils. Crude oils originating from diff erent producing zones, wells, or fi elds are mixed through commingling operation Geochemical-based oil allocation has been applied for for reducing operation cost, improving production allocation of commingled production from multiple zones effi ciency and fi eld management. For technical reasons, it is and production from multiple fi elds to a single stream [1, sometimes necessary to quantitatively deconvolute these 2]. This method uses whole-oil chromatography distinctive mixtures to facilitate monitoring of individual zones, wells, peak height ratios to quantify the contribution of each or fi elds. For example, where multiple pipelines commingle, individual production stream to a commingled production an accurate assessment of individual fi eld contributions stream. The distinctive compounds are natural fi ngerprints may be essential for establishing sales value or tax liability for oil classifi cation or oil grouping and used for proration because oil quality or tax rates can be diff erent between of each end-member oil in the commingled oil [3]. neighbouring fi elds that share a transport pipeline to 2. Methodology terminal facilities. The ability to back allocate helps in eff ectively managing reservoirs and production operations. The methods are based on the assumption that oils Normally, fi eld/reservoir allocation is done by from separated reservoirs or diff erent parts of a reservoir monitoring individual fi eld production with fl ow meters bear diff erent chemical signatures or distinctive chemical or by MPLT (Memory Production Logging Test) running. fi ngerprints. At the reservoir scale, steep gradients However, in this case, there was a lack of dedicated metering in oil composition and associated fl uid properties for production from some fi elds/reservoirs during a period are understood to be the product of preferential of time. Moreover, the tax liability was diff erent for those biodegradation of diff erent hydrocarbons, which gives neighbouring fi elds that shared a pipeline to transport oil oils a distinct molecular signature or “fi ngerprint” related to terminal facilities. Thus, it was necessary to investigate to the level of degradation. This natural variability in oil alternative methods for reallocating production from composition can be used to allocate oil production along these fi elds/reservoirs in order to determine an accurate horizontal wells or to assess the contribution of diff erent fi eld/reservoir contribution. The use of geochemical production streams in commingled wells by mapping the fi ngerprinting technique presented in this report is one original oil composition distribution.
Date of receipt: 6/9/2017. Date of review and editing: 6/9 - 14/9/2017. Date of approval: 4/10/2017.
PETROVIETNAM - JOURNAL VOL 10/2017 45 PETROLEUM EXPLORATION & PRODUCTION
Each oil type is identifi ed and clarifi ed through a data set oil due to continuous, subtle changes in of physicochemical properties named fi ngerprints. Depending the maturity of source rock and changes in on diff erent targets, oil fi ngerprints are determined by gas precisely which part of the source rock is in chromatography, biomarker analysis, IR, UV-VIS instruments, trace the oil window. Since no two compartments element, and isotope analysis which can lead to high capacity of oil are of identical geometry, and since no two type identifi cation. For example, the original oil type of gasoline and compartments have exactly the same fi lling oil spill can be determined by fi ngerprint analysis. history, it is diffi cult to achieve precisely the same homogenised composition in two Using whole oil gas chromatography fi ngerprints in order to separate compartments - even with oil from identify oil type is the most popular method which uses a set of the same source. peak height ratios to identify oil type. The height or area of a GC (gas chromatography) peak refl ects the concentration of a hydrocarbon - Processes that aff ect oil composition component of the oil. Due to approximately 1000 peaks in a whole oil after oil enters a reservoir (such as gas chromatogram, the distinctive peaks are usually chosen according biodegradation, water washing, evaporative to experience rather than their chemical nature. Therefore, many sets fractionation) do not operate to exactly the of GC fi ngerprints exist from diff erent laboratories. same extent in separate compartments. Depending on the oil fi eld, these compositional diff erences exist If two oil zones (1 and 2) were for one or more of the following 3 reasons [4]: commingled, the respective contributions of the zones to the commingled sample could - Oils from diff erent sources diff er in composition since oils be determined by identifying the chemical from diff erent source rocks have diff erent times of generation and/or diff erences between end-member oils, with diff erent migration paths. the end member being pure samples from - Oil, which a source rock generates at a given time, diff ers slightly core or cuttings samples from the production both from subsequently generated oil and previously generated well or nearby delineation wells to those of the produced oils (1 and 2). The combination of mass chromatograms as molecular fi ngerprint data and direct measurement of specifi c component abundances in oil samples are then made numerically. The data were then used to mathematically express the composition of the commingled oil in terms of contributions from respective end-member oils. The relationship between a GC peak height ratio Y (measured in the GC of a commingled oil) and the GC peak height ratio X of the corresponding peaks in the “m” end- Figure 1. A whole oil gas chromatogram [5] member oils being commingled is given by a linear relationship of the form:
Y = β1X1 + β2X2 + ... βmXm (1)
Mechanical P In which: Y: The peak height ratio matrix of Perforations ESP commingled oil;
X1, X2... Xm: The peak height ratio matrix of m end-member oils;
β1, β2... βm: The percentages of m end- member oils. Figure 2. Commingled production and allocation [6]
46 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Production allocation is the process of determining the values The problem thus becomes one of a linear of β1, β2... βm for end-member oils. regression. Specifi cally, given a set of samples containing a value for each independent variable In reality, there are multiple sources of error, including the and the corresponding value of the dependent following: variable, the “β” values must be computed in a - Analytical error in measuring the height of each GC peak; relationship of the form: - Error associated with potential contamination of GC peaks; Y = β1X1 + β2X2 + ... βmXm + E - Error associated with the non-ideality of the samples Where, “E” represents the error that is not chosen as end-members. captured by the linear relationship. Because of the various sources of error, certain GC peaks will Linear regression can be used to converge do a better job than others at allocating each zone’s contributions upon a set of β value that minimises the sum of to a commingled oil. It is impossible to know, in advance, which GC the square of errors (min|E|2). peaks will do the best job of the potentially hundreds available in an oil allocation process. 2.1. The advantage of geochemical allocation Geochemical allocation of commingled Input data Input p N. Input n_tols Input n_iter X (m-by-n) Sources Convg. Criteria Max. Iter. production from multiple zones in a single well typically costs less than 5% as much as conventional e-line production logging. The cost
Calculate || X - GF ||2 Initialize G & F saving is even more dramatic when compared 2 F := ref or F := random; eucl = || X - GF || G := random; to those for coiled-tubing-conveyed or tractor- conveyed MPLTs. The low cost of geochemical production Inner Loop allocation allows fi eld engineers to monitor Update F output frequently over long time periods (e.g., (GTX) F = F i- i- i- T weekly, monthly, and quarterly), allowing early (G GF)i- identifi cation of zone performance problems. The much higher cost of production logging Outer Loop limits that technique to infrequent use. Therefore, Update G (XFT) production logging typically provides only a G = F -j -j -j T T snapshot of the production origin at the time (G FF )-j the log was run, rather than a continuous performance history [2].
Yes i < p i = i +1 Geochemical fi ngerprinting techniques are applicable to highly deviated and horizontal No wells, in addition to vertical wells. By contrast, production logging interpretation is problematic Calculate || X - GF ||2 in highly deviated wells. Geochemical update_eucl = || X - GF ||2 fi ngerprinting techniques can be applied to all types of pumping wells, including those with tubing-deployed electrical submersible pumps,
i_step > n_iter or No i_step = i_step + 1 and progressive cavity pumps. Other than eucl - update_eucl < n_tols eucl := update_eucl those with unusual completion style, such as Y-block completion, most pumping well cannot Yes accommodate a production logging tool, because End NNMF the pumping apparatus prevents logging tool’s access to the underlying perforated interval. Figure 3. An example of NNMF (non-negative matrix factorisation) algorithm in least square technique [7]
PETROVIETNAM - JOURNAL VOL 10/2017 47 PETROLEUM EXPLORATION & PRODUCTION
The geochemical allocation approach can also be used to application of this method for commingled assess the contribution of multiple fi elds to commingled pipeline well P01 in A fi eld, off shore Vietnam. production streams. There are some additional advantages including 3.1. Sampling and analysis the ability to allocate in the absence of fl ow meter data, and the ability to identify problems with fl ow meter data. Miocene reservoir, A fi eld, includes 2 pay zones: clastic and carbonate. P01 well started 2.2. Procedure producing in 2011 and has been perforated at Oil samples collected from wellhead were taken to the laboratory both clastic and carbonate reservoirs. In order for geochemical analysis. GC processing was carried out by to determine the oil contribution of these two Chemstation software, chemometric tools and polynomial statistic reservoirs to P01 well, clastic and carbonate oil to group oils and identify distinctive pairs of peaks for allocation samples from P02 and P03 respectively were calculation. collected to be the end-members.
3. Case study of geochemical production allocation The whole oil samples (Table 1) were fully analysed on Agilent 6890N. Gas Geochemical-based allocation method was developed by the chromatograms are input data for Geochemical Vietnam Petroleum Institute and has been applied for hundreds of Allocation software to identify oil group and oil samples collected from diff erent fi elds. The paper introduces the allocate commingled production.
3.2. Qualitative results Thermostatic Oven Sample Flame Inlet Ionization Oil - oil correlation is usually evaluated Detector Increasing carbon 0.1μl number according to [1, 8]: FID - light hydrocarbon distribution (based Intensity
on C7 isomers);
Time - hierarchical cluster analysis (based on Carrier Gas Each peak represents whole oil gas chromatograms). Helium individual compounds or several compounds Light hydrocarbon distribution (B-F Figure 4. The principal diagram of a GC-FID system [5] diagram) was used to determine the diff erences between Miocene oils produced from the clastic and carbonate reservoirs of A fi eld. Figure 6 shows the correlation between aromatic index (toluene/n-heptane, B) and saturated index (n-heptane/methylcyclohexane, F). The diff erences between oils produced from two pay zones are displayed through alkane, aromatic and naphthene hydrocarbon correlation. The negative slope illustrates the secondary alternation such as evaporative fractionation, thermal cracking, water washing, and biodegradation [8]. When this slope translates to the right, the maturity of source Figure 5. Agilent 6890N gas chromatography rock generating these oils increases. Table 1. Oil samples collected from A fi eld Sample code Quantity Reservoirs Sampling point VPI-P01 16 Clastic + Carbonate Wellhead VPI-P02 12 Clastic Wellhead VPI-P03 10 Carbonate Wellhead
48 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
As shown in the B-F diagram, the 6.5 VPI-P01 samples of VPI-P02 and VPI-P03 are located 6 VPI-P02 at two separated areas, it means that the oil 5.5 VPI-P03 geochemical properties from these two wells 5 are highly diff erent (VPI-P02 represents oil
4.5 from clastic reservoir and VPI-P03 represents Clastic oil from carbonate reservoir). Moreover, the oil 4 trend samples from these VPI-P02 and VPI-P03 are 3.5 Carbonate Toluene/n-heptane trend located along two parallel straight lines which 3 therefore can be considered as the trends of 2.5 clastic and carbonate oils respectively. 2 The oil samples from VPI-P01 were found 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 very near the location of carbonate trend. n-heptane/methylcyclohexane These results illustrate that oil from carbonate Figure 6. B-F diagram of wells P01, P02 and P03 reservoir mostly contributes to the oil production stream of VPI-P01. Using all GC whole oil data of VPI-P01, VPI-P02 and VPI-P03 as an input data to run the cluster analysis, the dendrogram is shown in Figure 7 with two distinct groups. Accordingly, the samples of VPI-P01 and VPI-P03 are grouped in one cluster and the other cluster is VPI-P02 with very large distance. Hence, the oils from VPI-P01 and VPI-P03 have a distinct carbonate reservoir signature, whereas the oil from VPI-P02 has clastic signature. The distances between the groups (along the x-axis) are a relative measure of the magnitude of diff erences among the oils. Therefore, carbonate and clastic oils are clearly separable Figure 7. Hierarchical cluster analysis diagram of VPI-P01, VPI-P02 and VPI-P03 with large compositional discrepancy. Once again, this result proves that the oil fl ow from carbonate reservoir mainly contributes to the i700/i661 i1100/i1085 30 i700/i674 production stream of VPI-P01. i1065/i1049 i700/i677 P02 i1065/i1045 25 i700/i681 In addition, from over 500 20 i1025/i1038 i800/i784 chromatographic peaks (corresponding to 15 more than 500 hydrocarbon compounds in i1000/i1016 10 i800/i785 a GC chromatogram from C6 to C14), through i1000/i1008 5 i800/i790 the adaptive algorithm and visual inspection, P03 0 26 pairs of distinctive chemical compounds i1000/i1005 i800/i798 (or 26 distinctive peak height ratios) were i1000/i993 i900/i908 screened and chosen to display on the i976/i963 i900/i909 stardiagram. According to the star diagram,
i972/i984 i900/i883 P01 the diff erence between the geochemical i966/i970 i953/i963 properties of VPI-P01/VPI-P03 and VPI-P02 i963/i959 i955/i959 i963/i970 can be easily observed (Figure 8).
Figure 8. Polar (star) plot of selected peak ratios for well P01, P02, and P03
PETROVIETNAM - JOURNAL VOL 10/2017 49 PETROLEUM EXPLORATION & PRODUCTION
3.3. Quantitative results Proceedings of the 9th Annual Research Conference of the Society of Economic Paleontologists and Mineralogists, Assuming oil from well VPI-P02 is an end‐member New Orleans. 1 October, 1990. oil of the clastic reservoir and oil from well VPI-P03 is an end-member oil of carbonate reservoir, the oil from well 2. R.L.Kaufman, H.Dashti, C.S.Kabir, J.M.Pederson, VPI-P01 is now considered as commingled oil from two M.S.Moon, R.Quttainah, H.AI-Wael. Characterizing reservoirs and therefore need to be allocated. Distinctive the Greater Burgan fi eld: Use of geochemistry and oil peak height ratios are chosen from approximately 1,000 fi ngerprinting. SPE Reservoir Evaluation and Engineering. 2002; 5(3): p. 190 - 196. gas chromatography peaks from C6 to C35 as input data of the Geochemical Allocation Software for allocation 3. Barry Bennett, Jennifer J.Adams, Stephen R.Larter. calculation. Oil fi ngerprinting for production allocation: Exploiting the The allocation results of commingled well VPI-P01 show natural variations in fl uid properties encountered in heavy the contribution of both clastic and carbonate reservoirs to oil and oil sand reservoirs. Frontiers + Innovation. CSPG oil stream of P01 and vary from time to time. However, oil CSEG CWLS Convention. 2009: p. 157 - 160. from carbonate reservoir mainly contributes to commingled 4. Brooks A. Patterson and Mark A. Beeunas, oil stream of well P01 during the observed period. It Geochemical comparison of Rang Dong Field oils, Cuu Long illustrates that oil produced from P01 is characterised by basin, Vietnam. 2006. oil produced from carbonate reservoir. These results are 5. David K.Baskin, Alan S.Kornacki, Mark A.McCaff rey. consistent with the qualitative results in 3.2. Allocating the contribution of oil from Eagle Ford formation, In order to evaluate the accuracy and validate the the Buda formation, and the Austin Chalk to commingled technique, the geochemical allocation results were production from wells in South Texas using geochemical compared with fi eld production data based on MPLT fi ngerprinting technology. AAPG Annual Convention and running for well VPI-P01. Allocation of well VPI-P01 oils Exhibition, Pittsburgh, Pennsylvania. 19 - 22 May, 2013. based on logging result compares very favourably with the 6. Scott Ramos, Brian Rohrback, Glenn Johnson, geochemical result on the same date. The high agreement Russell Kaufman. Using gas chromatography and curve (in this case less than 1%) observed indicates the validity resolution to quantify contributions to mixed crude oils. 56th of the geochemical fi ngerprinting method. Pittsburgh Conference, Orlando, Florida. 2005. 4. Conclusion 7. Daniel D.Lee, H.Sebastian Seung. Algorithms for Geochemical fi ngerprinting technique with gas non-negative matrix factorization. Advances in neural chromatography is a valuable method to accurately information processing systems 13. The MIT Press. 2001: allocate commingled production stream and production p. 556 - 562. pipeline. The results of allocation calculation of 8. H.I.Halpern. Development and application of light- commingled wells based on their geochemical hydrocarbon based star diagrams. AAPG Bulletin. 1995; fi ngerprints are in excellent agreement (generally within 79(6): p. 801 - 815. 3%) with the actual production logging data. This method 9. Marcio M.Lobão, Jari N.Cardoso, Marcio R.Mello, can be used to replace other methods (MPLT or metering) Paul W.Brooks, Claudio C.Lopes, Rosangela S.C.Lopes. to determine the oil allocation from commingled wells/ Identifi cation of source of a marine oil-spill using geochemical pipelines more frequently because of their cost saving and chemometric techniques. Marine Pollution Bulletin. and high accuracy. 2010; 60(12): p. 2263 - 2274. References 10. Nguyen Xuan Thanh. Reservoir geochemical 1. R.I.Hwang, D.K.Baskin. Reservoir connectivity and evaluation of the Rang Dong basement oils in off shore oil homogeneity in a applications in the Gulf of Mexico. Vietnam. 1999.
50 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
TESTING ANTIBACTERIAL EFFECT OF SILVER NANOPARTICLES ON SULFATE-REDUCING BACTERIA Cu Thi Viet Nga, Trinh Thanh Son, Ha Thu Huong, Nguyen Thi Ngoc Bich, Ngo Hong Anh Vietnam Petroleum Institute Email: [email protected]
Summary
In the oil and gas industry, sulfate-reducing bacteria can generate hydrogen sulfide (H2S) in their growth, thus decreasing the commercial value of crude oil, causing metal equipment corrosion and seriously aff ecting the economic efficiency and the health of workers on drilling rigs. Because of their strong antibacterial and environmental friendly characteristics, silver nanoparticles have been studied and applied in various sectors world-wide. This paper presents some initial research results on the synthesis of silver nanoparticles (NPs) with the mean silver nanoparticle size of 18 - 21nm being eff ective to reduce bacterial numbers from 106tb/ml to < 10tb/ml (99%). This activity was tested on sulfate-reducing bacteria EPC-KK2 - Desulfomicrobium baculatum which was isolated from crude oil samples taken from Bach Ho field. Key words: Silver nanoparticle, biocide, sulfate-reducing bacteria, Bach Ho fi eld.
1. Introduction of silver NPs to free Ag+ and free ions that will aff ect bacteria. Some mechanisms of silver NPs action on The sulfate reducing bacteria (SRB) are anaerobic bacteria are being understood, for example silver ions bacteria that produce H S. This type of bacteria can survive 2 inhibit the ability of oxygen to travel through the cell wall in extreme environments such as high temperature, high due to its ability to bind to the peptidoglycan (cell-wall pressure, high salinity, to alkaline or acidic environment component), leading to cell death [4]. In animals without and is especially quite common in oil fi elds and cell walls, they are unaff ected by exposure to these ions. exploitation wells. The SRB produce H S that is acidifying, 2 In the other mechanism, silver ions can pass through the leading to a reduction in the commercial value of crude cell membrane into the cell and react with sulfhydryl - oil, corrosion of metalworking equipment, piping, and the SH group of the oxygen-converting enzyme molecule, health of workers on drilling rigs [1, 2]. Under favourable neutralising the enzyme that inhibits bacterial cell conditions, desulfurising bacteria can develop biofi lm respiration. In addition, silver ions are capable of binding membranes that occlude the reservoir, which reduces the to DNA bases and neutralising the charge of phosphate, ability of the water to pump. preventing DNA replication [5 - 8]. Currently, popular biocides being used are aldehydes Antibacteria with silver NPs is a new research direction or cyclic amines in combination with cationic active in the manufacture of biocides for use in the oil and gas substances, which are highly toxic to humans and the industry. This article presents some results of synthesis environment. When these biocides are used for a long time the phenomenon of “grease” will occur. It is therefore of silver NPs and testing bactericidal ability of silver NPs in need of a new biocide with better antibacterial eff ect on desulfurising bacteria isolated from Bach Ho crude oil and being more environmentally friendly. samples. Silver NPs is a new material that is known for its 2. Materials and methods superior disinfection ability and has many applications in 2.1. Materials and equipment life. Silver NPs are very small in size and have very large surface area. They can penetrate easily into microbial 2.1.1. Materials cells, alter biochemical mechanisms, inactivate microbial In this study, all reagents of producing silver NPs metabolism, and fi nally destroy them [3]. colloids were of analytical grades and used as received
The antibacterial mechanism of silver NPs is the without further purifi cation. Silver nitrate AgNO3 (99.98%) result of the conversion of silver atoms on the surface was used as the silver precursor, which was obtained from
Date of receipt: 15/10/2017. Date of review and editing: 15/10 - 25/10/2017. Date of approval: 26/10/2017.
PETROVIETNAM - JOURNAL VOL 10/2017 51 PETROLEUM PROCESSING
Merck (Germany). Sodium borohydride NaBH4 (96.5%) Evaluating the dispersibility of silver NPs: Silver NPs were was obtained from Chemical Ltd., whereas the sodium diluted at diff erent concentrations and UV-VIS spectra were citrate Na3C6H5O7 (99.0%) from Sigma Aldrich was measured to evaluate the existence and contraction of silver used as producing agents. All the aqueous solutions NPs. were prepared in double distilled water. Isolation and identifi cation of SRB: Bacteria were isolated The improved Postgate’s B bacterial culture from Bach Ho oil samples. SRB was isolated on Postgate’s B medium was used for bacterial culture KH2PO4 0.5g/l; medium and purifi ed to the mono-assay by means of critical
Na2SO4 1g/l; NH4Cl 1g/l; MgSO4 2g/l; NaCl 4g/l; CaCl2 dilution. SRB strain was Gram stained and its cell shape was 0.1g/l; marine water 200ml; distilled water 800ml; observed under the optical microscope. C H NaO 2g/l; pH 7.2g/l; C H NaO .3H O 1.75g/l; 3 5 3 2 3 2 2 The bacterial strain was cultured and enriched, and FeSO 0.5g/l; yeast extract 0.5g/l; the medium should 4 identifi ed by analysing and sequencing of 16S rRNA genes. be autoclaved for 15 minutes at 121oC (250oF). The The biocide effi ciency was tested according to bacterial pH should be adjusted with NaHCO3. growth after exposure to silver NPs according to API RP38 SRB strain was isolated from MSP10 crude oil standard. sample of Bach Ho fi eld and identifi ed by analysing and sequencing 16S rRNA genes. 3. Results and discussion
2.1.2. Equipments 3.1 Preparation of silver nanoparticle solution 3.1.1. Selection of reducing agents The UV-visible spectra were recorded over the range of 200 - 800nm by Thermo Evolution 300. The The results in Figures 2 and 3 show that the particle sizes of size and morphology of silver nanoparticle colloids were investigated using transmitted electron Stabilising AgNO3 microscope (TEM) Philips CM 120. Particle size measuring device by laser light scattering LA-950. Drop slowly Reducing agent Deionised (B solution) 2.2. Methods
A solution The chemical reduction method is used for the production of silver NPs from AgNO , with the 3 Silver nanoparticles reducing agent NaBH4 or Na3C6H5O7 [9]. The reduction solution process for silver NPs is shown in Figure 1. Figure 1. Diagram of silver nanoparticle preparation
Figure 2. UV-VIS spectrophotometric and TEM image of silver NPs with Na3Cit
52 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Figure 3. UV-VIS spectrum and TEM image of silver NPs with NaBH4
Figure 4. UV-VIS spectra of silver nanoparticle samples (AgNPs 30, 31, 33) Figure 5. UV-VIS spectra of silver NPs solution samples with concentration ratios
stabilised/AgNO3: 0.05; 0.275 and 0.5 products obtained from Na3Cit (Na3C6H5O7) reductant are comparing to AgNO3. Figure 5 shows the UV-VIS larger than those from NaBH4. When Na3Cit was used, the spectrographic chromatograms of silver NPs colloids with absorption wavelength of the silver NPs colloid went up to NaBH4 to AgNO3 ratios being 2:1; 3:1; 4:1 at 0.380; 0.900; 450nm under UV-VIS spectrometer and the effi ciency of 0.683 peak heights respectively. It means that the content reaction reduced signifi cantly comparing to NaBH4. of reducing agents would considerably aff ect the numbers It means that the reduction capacity of Ag+ ions to of silver NPs produced (Figure 4). The results showed o that the highest concentration of silver NPs colloid was Ag by NaBH4 is higher than Na3Cit. As a result, NaBH4 was obtained at NaBH to AgNO ratio of 3:1. chosen for the synthesis of silver NPs. 4 3 - Eff ect of the stabilising agent/AgNO ratio 3.1.2. Silver nanoparticle preparation conditions 3 The results in Figure 5 indicated that the UV-VIS - Eff ect of NaBH /AgNO concentration ratio 4 3 absorption peak height of products obtained from a 0.05
In order to achieve the highest reaction effi ciency, ratio of stabilising agent to AgNO3 is much lower than the NaBH4 reductant must be used in excess amount that from 0.275 and 0.5 ratios. It can be explained that
PETROVIETNAM - JOURNAL VOL 10/2017 53 PETROLEUM PROCESSING
too low stabilising agent concentration will not create mixed molecular surrounding silver NPs, and this is the reason why the nanoparticle size increases and the number of silver NPs decreases. The highest UV-VIS absorption peak was obtained from the 0.5 ratio of stabilising
agent to AgNO3. 3.1.3. Evaluation of silver NPs colloid characteristics
Figure 6. Silver NPs solution in diff erent concentrations Silver NPs colloids were prepared at diff erent concentrations from 0.2 to 2,000ppm (Figure 6). UV-VIS absorption spectrometer of range of silver NPs concentrations was shown in Figure 7. The diluted samples all reached a peak of 400nm which indicated that the properties of silver NPs were not aff ected by the dilution in water. The TEM image of silver NPs at 2,000ppm (Figure 8) showed that the size of silver NPs was at approximately 15 ± 4nm. These nanoparticles meet the requirements of size which is eff ective for biocide (< 100nm). Figure 9 showed the results of size distribution of silver NPs at concentration, 2000ppm, and the mean size is around 19nm.
3.2. Isolation and identifi cation of bacteria Figure 7. UV-VIS spectra of silver nanoparticles AgNPs 27 at concentrations of 0.1%, 0.3%, 0.5%, 1% and 1.5% The EPC-KK2 sulfate reducing strain was isolated from oil sample MSP10-1014. On the microscope, these cells are short rod shaped, Gram-negative bacteria, planktonic. This type of bacteria is anaerobic, and anaerobic respiration uses sulfate as the fi nal electron acceptor,
producing H2S. They grow on lactate substrates. Optimum conditions of growth include: appropriate temperature is 30 - 32oC; pH is 7.5; and salt concentration of NaCl is 2.5g/l. The EPC-KK2 strain was classifi ed and identifi ed by analysing and sequencing 16S rRNA genes. The results of gene analysis (Figure 10) of EPC-KK2 showed that the amplifi ed genomes were similar to Desulfomicrobium baculatum KF536747.1 (Figure 11). From the results of sequencing 16S rRNA genes and the cell morphology (Figure 12), EPC-KK2 can be Figure 8. TEM of AgNPs 32
54 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
50 100 identifi ed as Bacteria; Proteobacteria; Deltaproteobacteria; Desulfovibrionales; 45 90 Desulfomicrobiaceae; Desulfomicrobium 40 80 baculatum. 35 70 30 60 General characteristics of the 25 50 bacterium Desulfomicrobium baculatum: q (%) 20 40 Cells: rod or ellipsoidal-shaped Undersize (%) 15 30 cells, 0.5 - 0.9 x 1.3 - 2.9μm, with round 10 20 ends, either singly or in pairs [10]. Gram- 5 10 negative stain reaction and having cell- 0 0 wall structure. Cells are motile, usually 0.01 0.1 1 10 100 1000 by a single polar fl agellum. Endospores Diameter (μm) are not formed. Figure 9. Size distribution of AgNPs 27 in solution Growth condition: anaerobic, pre- AGTCAATGAACGCTGGCGGCGTGCCTAACACATGCAAGTCGTACGAGAAAGGGGCTTTCGGGCCCTGAGT reduced medium or reducing agent AGAGTGGCGCACGGGTGAGTAACGCGTGGGTAATCTACCCTTGGATTTGGGATAACTCTGCGAAAGTGGA GCTAATACCGGATAGTCTGGCTTTAATTAAGAAGTCGGTAAAGGATGCCTCTGCATATGCATTCGTCCGA required in medium for growth. Growth GGATGAGCCCGCGTCTCATTAGCTAGTTGGTAGGGTAATGGCCTACCAAGGCAACGATGAGTAGCTGGTC can occur by anaerobic respiration with TGAGAGGATGATCAGCCACACTGGAACTGAAACACGGTCCAGACTCCTACGGGAGGCAGCAGTGGGGAAT sulfate or sulfoxy-anions as terminal ATTGCGCAATGGGCGAAAGCCTGACGCAGCAACGCCGTGTGAGGGATGAAGGCTTTCGGGTCGTAAACCT CTGTCGGAAGGGAAGAACGGGCATTGGTCTAATAGGCCTTTGTTTTGACGGTACCTTTAGAGGAAGCACC electron acceptor, producing H2S. GGCTAACTCCGTGCCAGCAGCCGCGGTAATACGGAGGGTGCAAGCGTTATTCGGAATTACTGGGCGTAAA Optimal temperature from 25 - 30oC. Figure 10. Sequence genes of strain EPC-KK2 Substrates used: simple organic
Desulfovibrio desulfuricans AM949034.1 compounds that serve as electron donors during sulfate respiration include Bacillus megaterium EU124555.1 lactate, pyruvate, ethanol, formate and hydrogen. Sulfate respiration with Azospirillium brasilense NR_117478.1 lactate as electron donors is incomplete,
with the formation of acetate and CO2. Hydrogenase is present. Cells contain 1 b- and c-type cytochromes. Metabolism
Desulfovibrio desulfuricans KJ459864.1 can also be fermentative on simple organic substrates, including pyruvate, malate or fumarate; carbohydrates are not fermented. No specifi c vitamins are 2 EPC -KK2 required [10]. Desulfomicrobium baculatum KF536747.1 3.3. Antibacterial eff ect of silver NPs Desulfomicrobium baculatum AJ277895.1 There are a number of theories on Desulfomicrobium baculatum AM419440.1 the antibacterial ability of silver NPs colloid, in which the absorbed theory is most convincing. The main idea of the theory is that the bacterial cell is inactive Desulfobulbus elongatus NR_029305.1 due to the electron storage between the 3 negative charged surface of the cell and the Ag+ ions which have been absorbed Desulfobulbus elongatus X95180.1 to it. These ions would penetrate into the 0.1 inside of the bacterial cell and inactivate Figure 11. Phylogenetic position of EPC-KK2 species with relatives based on16S rRNA sequences them.
PETROVIETNAM - JOURNAL VOL 10/2017 55 PETROLEUM PROCESSING
Table 1. The eff ect of NPs size and concentration of AgNPs colloids on EPC-KK2
Original Number of Number of Concentration EPC-KK2 after EPC-KK2 after Mean particle number Control No. Sample of silver NPs 24 hours in 96 hours in size (nm) of EPC-KK2 K (ppm) silver NPs silver NPs EPC-KK2 (cell/ml) (cell/ml) (cell/ml) 1 AgNPs 25_30 21.45 30 106 nd nd 106 2 AgNPs 25_60 21.45 60 106 nd nd 106 3 AgNPs 25_100 21.45 100 106 nd nd 106 4 AgNPs 27_30 18.45 30 106 nd nd 106 5 AgNPs 27_60 18.45 60 106 nd nd 106 6 AgNPs 27_100 18.45 100 106 nd nd 106 7 AgNPs 31_30 25.28 30 106 105 105 106 8 AgNPs 31_60 25.28 60 106 105 104 106 9 AgNPs 31_100 25.28 100 106 106 105 106 10 AgNPs 34_30 25.38 30 106 105 105 106 11 AgNPs 34_60 25.38 60 106 105 105 106 12 AgNPs 34_100 25.38 100 106 106 105 106 nd: not detected Table 2. Test on antibacterial concentrations of AgNPs 25 with EPC-KK2 Concentration of Number of EPC-KK2 Number of EPC-KK2 Original number of Control No. silver NPs in solution after 24 hours in after 96 hours in EPC-KK2 (cell/ml) K (ppm) silver NPs (cell/ml) silver NPs (cell/ml) EPC-KK2 1 0 106 10 6 10 5 10 6 2 5 106 10 6 10 5 10 6 3 10 10 6 10 5 10 4 10 6 4 15 10 6 10 4 10 3 10 6 5 20 10 6 ~10 nd 10 6 6 25 10 6 nd nd 10 6 7 30 10 6 nd nd 10 6 nd: not detected The silver NPs samples collected are tested with the antibacterial eff ect based on their particle sizes. According to announced studies, the size of silver NPs is an important factor that aff ects the antibacterial property. For diff erent size of silver NPs, the activities of the solution on the bacteria are very diff erent [8]. Overall, the size of the silver NPs that is eff ective for bacteria is within the range of 1 to 50nm. For the isolated strain, the size of silver NPs is within 1 - 50nm. The eff ective concentration of silver NPs solution on diff erent bacteria varies. Normally, the minimum inhibited concentration of silver NPs solution is about 10 - 75ppm. The silver NPs solutions which are prepared and tested in the study include: AgNPs 25, AgNPs 27, AgNPs 31, and AgNPs 34. Figure 12. Sulfate reducing bacteria EPC-KK2 under Leica DM750 microscope These solutions did not show precipitation and were tested with the EPC-KK2 type. The results are shown in mean particle size > 25nm did not kill EPC-KK2 strain, Table 1. eff ectively. Table 1 shows the ability of silver NPs colloid AgNPs With the silver NPs size < 25nm, the antibacterial 25 and AgNPs 27 to kill EPC-KK2 strain with the mean eff ect of AgNPs 25 and AgNPs 27 were observed at particle size < 25nm. AgNPs 31 and AgNPs 34 with the diff erent concentration between 0 - 30ppm. The results
56 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Table 3. Test on antibacterial concentrations of AgNPs 27 with EPC-KK2 Number of EPC - Concentration of Number of EPC-KK2 Original number of KK2 after 96 hours Control No. silver NPs in after 24 hours in silver EPC-KK2 (cell/ml) in silver NPs K solution (ppm) NPs solution (cell/ml) EPC-KK2 solution (cell/ml) 1 0 106 106 106 106 2 5 106 106 106 106 3 10 106 104 104 106 4 15 106 104 103 106 5 20 106 ~10 nd 106 6 25 106 nd nd 106 7 30 106 nd nd 106 nd: not detected in Tables 2 and 3 show that EPC-KK2 was not observed after the experiment with AgNPs 25 and AgNPs 27 samples at the concentration of 20ppm. Pictures of the cultivated EPC-KK2 strain in the experiment to assess the antibacterial ability of silver NPs colloid are shown in Figures 13 and 14. The results of the tests show that the antibacterial ability of AgNPs 27 and AgNPs 25 samples are highly eff ective at the concentration of 20ppm for the EPC- KK2 strain.
4. Conclusion
The EPC-KK2 strain, which was isolated Figure 13. EPC-KK2 strain grown again in test of silver NPs from the Bach Ho crude oil samples and classifi ed as Desulfomicrobium baculatum, was used to evaluate the antibacterial eff ectiveness of silver NPs colloid. Silver NPs (mean silver NPs size 18 - 21nm) were eff ective to reduce the antibacterial activity from 106tb/ml to < 10tb/ml (99%) at 20ppm. The results showed that silver NPs are highly eff ective in killing SRB. This would create a new approach for applying silver NPs as biocide in Vietnam's oil and gas industry.
References
1. Lai Thuy Hien, Dang Phuong Nga. Physiological and biochemical properties of some sulfate-reducing bacteria strains isolated from Bach Ho oil fi eld. Journal of
Figure 14. EPC-KK2 strain killed in test of silver NPs Biology. 1998; 20(2): p. 33 - 38.
PETROVIETNAM - JOURNAL VOL 10/2017 57 PETROLEUM PROCESSING
2. Lai Thuy Hien, Le Phi Nga. Study on 7. Reed M.Izatt, James J.Christensen, J.Howard metal corrosion potential of Desulfovibrio Rytting. Sites and thermodynamic quantities associated vulgaris. Journal of Biology. 1992; 14(4): with proton and metal ion interaction with ribonucleic p. 26 - 29. acid, deoxyribonucleic acid, and their constituent bases, nucleosides, and nucleotides. Chemical Reviews. 1971; 3. Catalina Marambio-Jones, Eric M.V.Hoek. A 71(5): p. 439 - 481. review of the antibacterial eff ects of silver nanomaterials and potential implications for human health and the 8. Gianluigi Franci, Annarita Falanga, Stefania environment. Journal of Nanoparticle Research. 2010; Galdiero, Luciana Palomba, Mahendra Rai, Giancarlo 12(5): p. 1531 - 1551. Morelli, Massimiliano Galdiero. Silver nanoparticles as potential antibacterial agents. Molecules. 2015; 20(5): 4. T.A.Brown, D.G.Smith. The eff ects of silver nitrate p. 8856 - 8874. on the growth and ultrastructure of the yeast Cryptococcus albidus. Microbios Letters. 1976; 3: p. 155 - 162. 9. Shuai He, Honglin Chen, Zanru Guo, Biqing Wang, Chongli Tang, Yujun Feng. High-concentration silver colloid 5. M.R.Richards, H.A.Odelola, B.Anderson. Eff ect of stabilized by a cationic geminisurfactant. Colloids and silver on whole cells and spheroplasts of a silver resistant Surfaces A: Physicochemical and Engineering Aspects. Pseudomonas aeruginosa. Microbios. 1984; 39(157 - 158): 2013; 429: p. 98 - 105. p. 151 - 157. 10. George M.Garrity. Bergey’s manual of systematic 6. Yakabe Yoshikuni, Sano Takayuki, Ushio Hidetoshi, bacteriology (2nd edition). Springer Science & Business Yasunaga Tatsuya. Kinetic studies of the interaction between Media, USA. 2001; 2: p. 944 - 945. silver ion and deoxyribonucleic acid. Chemistry Letters. 1980; 9(4): p. 373 - 376.
58 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
ENERGY EFFICIENCY IN VIETNAMESE ENTERPRISES: THE PREDOMINANCE OF GAS CONSUMERS Nguyen Thanh Luan1, Pham Thi Thu Ha2 1Vietnam Petroleum Institute 2Hanoi University of Science and Technology Email: [email protected]
Summary
Energy ineffi ciency is a serious issue in the Vietnamese industry, which is the sector consuming the most energy in Vietnam. This article adopts stochastic frontier analysis (SFA) to evaluate the factors that infl uence the energy ineffi ciency in Vietnamese enterprises. Under the evaluation for the tile industry, the empirical results indicate that there are many factors having signifi cant impacts on energy ineffi ciency, including fuel type, ownership structure, size of fi rms, labour quality, energy price and import activities. Relatedto the fuel type indicator, this article discovers that the level of energy ineffi ciency in fi rms using gases is 1.258 times lower than in fi rms using other fuels. That is the obvious evidence for the better performance of gas consumers in term of energy effi ciency. Key words: Energy effi ciency, stochastic frontier, SFA, gas consumers, Vietnamese industrial sector, industry energy effi ciency.
1. Introduction According to the industrialisation strategy of the Vietnamese Government, it is expected that the The industrial sector plays an important role in the Vietnamese economy and the energy consumption development of the Vietnamese economy, contributing the will maintain similarly rapid growth rates up to 2020. largest share (38%) of the country’s GDP [1]. Corresponding Therefore, a possible consequence is that Vietnam may to the contribution rate of GDP, the industrial sector experience energy shortage soon or later. Reduction of accounts for the highest proportion of energy consumption the energy ineffi ciency, which is claimed as the national (38.2%) in total energy demand [2]. Nevertheless, one can target by the Vietnamese government, becomes an see that the ineffi cient energy usage in the industrial sector urgent mission. It is necessary to focus on minimising the is one of the main reasons for overall energy ineffi ciency in energy ineffi ciency in the industrial sector, especially in Vietnam. Table 1 implies that, to generate USD 1,000 of GDP, industries consuming high energy, such as tile production, the Vietnamese industrial sector consumes the highest steel industry, or cement production. Hence, the goals of amount of energy in comparison with many countries. More this article are to evaluate the energy ineffi ciency and to precisely, energy use per GDP of the Vietnamese industry identify the potential factors causing energy ineffi ciency is 3 times higher than the average energy consumption of in the industrial sector. The approach and the results of the world. It is also 1.7 times higher than the average level this article also can help the Government to fi nd better of other developing countries, including Thailand, China, strategies for improving energy effi ciency. Malaysia and Indonesia.
Table 1. Comparison of energy consumption per GDP by industry 2000 2010 2014 Vietnam 139.70 151.06 136.73 Thailand 124.94 130.07 132.86 China 175.20 145.04 134.39 Malaysia 155.11 138.42 128.15 Indonesia 126.74 103.70 88.35 Japan 126.85 113.10 97.64 United States 244.41 207.30 177.53 World 129.28 123.67 122.65 Unit: Kilogram oil equivalent per USD1,000. Source: WB, IEA (2017)
Date of receipt: 29/9/2017. Date of review and editing: 29/9 - 16/10/2017. Date of approval: 24/10/2017.
PETROVIETNAM - JOURNAL VOL 10/2017 59 PETROLEUM ECONOMICS & MANAGEMENT
Furthermore, the Vietnamese government is pursuing function can be written in the form of F(X, Y) = 0, where the goal of sustainable development, in which economic X is an M dimension non-negative input vector and Y is development must be accompanied by energy security an N dimension non-negative output vector. In case of and environment protection. It implies that cleaner fuels production with only one output, the production function like LPG or NG are more encouraged than the other types. will be Y = f(X1, X2, ..., Xn) where the function f(.) represents Thus, this article also aims to evaluate the diff erence in a technology governing the production process. energy effi ciency between companies consuming gases Based on technology of production, there are several (including LPG and NG) and those consuming other fuels. types of production function adopted in economic analysis, To the best of our knowledge, there is hardly any in which Cobb-Douglas and Translog production functions published study to assess the factors aff ecting energy are the most well-known. The Cobb-Douglass function effi ciency in the Vietnamese industry. Therefore, this form is described in normal form as 1 article is a signifi cant contribution to an overlooked or in natural logarithm form as 0 1 , 0 1 research fi eld in energy economics in Vietnam. In addition, where the parameters i satisfy that and 0 empirical researches assessing energy effi ciency can be 1 1 for all 1 , and 0 0 (A classifi ed into three main approaches: energy intensity, is a constant number). The Translog function form can be 1 data envelopment analysis (DEA) and stochastic frontier explained as 0 1 2 1 , analysis (SFA). While there are numerous studies applying where Y is output of production, Xi is inputs of production, energy intensity or DEA to measure energy ineffi ciency, are parameters and . there are only few studies applying SFA and all mainly focus In the literature of production function, the relation on country/region level. Therefore, another important between inputs and output is assumed as the frontier of contribution of this article is to explore the performance attainable production set. It means, with given inputs X , of SFA in energy economics with fi rm-level data. i a fi rm always produces the maximum possible output.
The remaining parts of this article are organised as Meanwhile, given output Y and all other inputs Xj (where follows. Section 2 describes the methodology employed j ≠ i), the value of input Xi is always determined at the in this article. Section 3 explains the estimation and the minimum level. However, in the production effi ciency data/variables collected to implement the model. Section literature, that assumption will be relaxed. Thus, with 4 provides the quantitative analysis. Section 5 discusses given inputs, fi rms may produce less than the maximum the policy implications of the empirical results. We possible output(s), or fi rms may need more than the summarise our main fi ndings in Section 6. necessary level of inputs to produce the given output. In those cases, fi rms have technical ineffi ciency. 2. Analytical framework Figure 1 provides a graphic demonstration of a Boyd [3] points out SFA can evaluate energy effi ciency technically ineffi cient production. Point A that locates either through a normal statistic function or through under production frontier Y = f(X) represents that a fi rm is production function. The former application is adopted to not technically effi cient. The ineffi ciency of fi rm A can be evaluate energy effi ciency for OECD countries and for US explained in two ways based on the input-oriented or the residents by Filippini and Hunt [4, 5]. The later approach, so- output-oriented measure. From the view of input-oriented called energy input distance function, is employed by Zhou, measure, the observed output level (Y0) can be technically Ang and Zhou [6], Honma and Hu [7] and Lin and Long [8]. produced consuming fewer inputs (X* instead of X0). In the Following Zhou, Ang and Zhou [6] and Lin and Long [8], graph, fi rm A can become effi cient by reducing inputs and this article employs the energy input distance function move to point C on production frontier. The distance AC approach with the assumption of Translog function. indicates the wasted inputs due to technical ineffi ciency and it is the basis to measure input-oriented (IO) technical 2.1. Technology ineffi ciency and input distance function ineffi ciency. More specifi cally, IO technical ineffi ciency is All production processes are understood as measured by which is the proportion of inputs processes to transform inputs into outputs under specifi c that fi rms can reduce to become effi cient. From the view technologies. The transformation relationship between of output-oriented measure, there is a possibility to inputs and outputs can be described by a mathematical technically attain higher level of output (Y* instead of Y0) formula, so-called the production function. A production for given inputs (X0), so that fi rm A can move to point B
60 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Y (4)
Y = f(X) It is possible to expand equation (4) with assumption B of Cobb-Douglas production function form or Translog Y* production function form as in equation (5) and equation (6), respectively. C 1 Y0 A 0 (5) 0
1 0 (6) O X* X0 X Figure 1. Input-oriented and output-oriented technical effi ciency Where v is statistical error, is measured by the fraction on the production frontier. The gap AB in the graph between total capital, K and energy consumption, E. Variable implies the output loses out because of technical is measured by the fraction between total labour, L and ineffi ciency and it is the basis to measure output- energy consumption, E. The non-negative energy ineffi ciency, oriented (OO) technical ineffi ciency. More precisely, u is measured in log form of input distance function u Ξ lnD (Y, OO technical ineffi ciency is measured by E E E K, L; E), where the higher value of u leads the lower value of ln(1/ which is the proportion of output that fi rms can E E ) and it is corresponding to the higher energy consumption, E. increase to become effi cient. it Thereafter, energy ineffi ciency can be determined by In mathematics, the input-oriented ineffi ciency adopting standard SFA estimation methods to estimate can be explained as l = Y . equations (5) and (6). Furthermore, this article also evaluates In that equation, u is input-oriented ineffi ciency, X the factors aff ecting energy ineffi ciency of fi rms through the D (.) ≥ 1 is the input distance function, homogenous l equation u = u(z ) where z is a vector of exogenous variables of degree one of inputs vector x, A(Y) is the required Eit it it such as energy price, the enterprise size or the ownership set of inputs that is feasible to produce the observed structure. Both of above goals can be achieved simultaneously outputs Y. Therefore, a fi rm will be considered as by the one-step estimation approach that is widely applied in effi cient if value of D is equal to 1. l recent researches. The detail of the estimation method will be 2.2. Energy input distance function described as in the next section.
Let E is vector of energy demand to produce 3. Estimation and data observed vector of outputs Y for given vector of non- 3.1. Estimation energy inputs X, the energy input distance function is stated that: Since the development of the estimation method for panel data by Pitt and Lee [9], the literature of SFA has steadily improved with several methods. This article follows the procedure of (1) Battese and Coelli [10], which is the most adopted method Choose =E, from (1) we can derive: in empirical researches, to evaluate the energy ineffi ciency via energy distance function. In addition, this article follows (2) suggestion of Lin and Long [8] to setup model with assumption of Translog production function form. The empirical model can More specifi cally, for a general production with be expressed as: two non-energy inputs being capital (K) and labor (L), 0 take a natural logarithm for both sides of equation (7) (2) we have: (3) (8) Let as energy ineffi ciency 0 in log form of input distance function, equation (3) (9) can be re-written:
PETROVIETNAM - JOURNAL VOL 10/2017 61 PETROLEUM ECONOMICS & MANAGEMENT
0 Firms’ total capital are collected as variable K, (10) which is measured in billion Vietnamese dongs (billion VND) based on the year 2010. Variable k, then, In this article, the Translog setting will be adopted Equation is derived by fraction between K and E, before taking (7) is energy frontier equation that is identical with equation natural logarithm to fi t with setup of the model (6). is energy ineffi ciency, is mentioned above. the statistical error. Equation (8) is ineffi ciency eff ects equation Total labour that participates in the production where zit is vector of exogenous variables (will be clarifi ed in process is denoted by L and is presented in number Section 3.2). is random variable truncated at of employees. We take the fraction between L and E point z , where is vector of parameters which may include to derive variable l, then take natural logarithm to be constant term. Equation (9) is to control heteroscedasticity in suitable with the setup of the SFA model. , where is variance of . Equation (10) is to control The value added of tile production is collected as heteroscedasticity in , where is variance of . Variable t variable Y, before taking natural logarithm to fi t with is time variable (if any) and are vectors of parameters. The the setup of the SFA model. Variable is measured in parameters of all equations are estimated simultaneously by billion Vietnamese Dong (billion VND)i. programme developed by Kumbhakar, Wang, and Horncastle [11] on the platform of Stata software. The following discussions will explain the potential factors which might aff ect the energy 3.2. Data ineffi ciency in the Vietnamese tile industry, which In order to provide an in-depth analysis and based includes fuel type, ownership structure, enterprise on availability of the data, we narrow down the scope of size, price of energy, labour quality and import the study focusing on the tile industry for our empirical activity. analysis. The data we used in this article is the latest fi rm- 3.2.2. Fuel type level panel data of the Vietnamese tile industry, which covers 574 fi rms throughout the territory of Vietnam over the The most important purpose of this article is period of three years. The entire data is collected from the to determine the diff erence in energy effi ciency Business Survey conducted by the General Statistics Offi ce between fi rms using conventional fuels (e.g. coal, of Vietnam annually. For the estimation of energy frontier diesel) and fi rms using cleaner energies (e.g. LPG or equations, we use four panel series which are: (1) aggregate NG). It is an well-known fact that use of gases leads to energy consumption, (2) total capital, (3) total labour that higher performance than use of other fuels. However, participates in the production process, and (4) the value under the same conditions of the production (same added. Meanwhile, for ineffi ciency eff ects equations we amount of output, same enterprise size, and quality include six variables: (1) fuel type, (2) ownership structure, of labour, etc) the diff erence in energy effi ciency of (3) enterprise size, (4) average price of energy, (5) quality of using gas and of using other fuels is still ambiguous. labour, and (6) import activity. In order to assess the eff ect of fuel type on energy 3.2.1. Variables in energy frontier equation ineffi ciency, this article adopts the dummy variable type, where type equal to 0 represents fi rms using Production process in tile companies consumes assorted gases and type equal to 1 is for production without energies including electricity, coal, gasoline, diesel, LPG and gases. natural gas. In addition, the combination of energies varies from fi rm to fi rm. Therefore, we adopt the widely accepted 3.2.3. Ownership structure approach that aggregates energy consumption which is the Many tile companies have state-owned share summation of the heating value of all consumed energies in capital, and it is understood that state-owned for producing purpose. Aggregate energy consumption, E, is fi rms are associated with ineffi ciency [12]. In order measured in billion British thermal unit (billion Btu). to capture that impact, this article uses the dummy
i There are 3 out of 1,722 observations in data set which have negative or zero value added and will be automatically excluded when we take log. Since they are very small in number, they do not aff ect the results of estimation.
62 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
variable gov_sh as an exogenous, where state-related fi rm dongs per person and they are transformed into natural has gov_sh equal to 1 and otherwise, gov_sh is equal to 0. logarithm form as lql.
3.2.4. Enterprise size 3.2.7. Import activity
The large-sized fi rms are usually considered more It is believed that fi rms having to import materials/ effi cient than smaller ones since they are able to purchase fuels will have more incentive to control its production better equipment and to recruit more productive to be more effi cient than other fi rms. This article creates a employees [8]. However, since large-sized fi rms have dummy variable im to investigate the diff erence in energy more advantages to access energy resources, they may ineffi ciency between fi rms having import activities (im have less incentive to consume energies effi ciently than equals to 1) and others (im equals to 0). smaller ones. This article evaluates the diff erence in Statistical summary of the above variables is given in energy effi ciency between diff erent sizes of fi rms through Table 2. dummy variable size. The large-sized fi rms (represented by size equal to 0) and small and medium-sized fi rms 4. Quantitative analysis (SMEs, represented by size equal to 1) are defi ned in the The estimation results displayed in Table 3 indicate that Vietnamese Government’s Decree No. 56/2009/ND-CP. all models are valid with high signifi cance in parameters. 3.2.5. Average price of energy In energy frontier equation, the estimated coeffi cients are strongly signifi cant at the 0.1% signifi cance level. Besides The price of energy is an important factor in energy that, in ineffi ciency eff ects equation, almost estimated effi ciency analysis, where the higher the energy price is, coeffi cients are highly signifi cant at level of 5% or even at the higher energy effi ciency might be. Due to the facts level of 0.1%. There is only one exception - the coeffi cient of that tile fi rms use many types of energies simultaneously im is only signifi cant at the level of 27% in case of Translog and there is no competitive energy market in Vietnam, function. Due to most of materials for the tile industry can the use of unit price becomes inappropriate. Thus, this be supplied by the domestic market, not many companies article employs the average price that is calculated by the have import activities. Thus, the signifi cance level of 27% division of total cost for energies and aggregate energy and the appropriate sign of coeffi cient (negative sign) are consumption determined before. Thereafter, they will be noticeable results to convince that we should not drop it transformed to natural log form and be denoted as lp. out of model. 3.2.6. Quality of labour Table 3 also shows that there is heteroscedasticity in Economists accept that the quality of labour is ineffi ciency term but no heteroscedasticity in statistical conducive to improving production effi ciency. This error. While a few coeffi cients in variance equation article employs the average salary, lql, to investigate the of ineffi ciency term are not really signifi cant, other impact of labour quality on energy ineffi ciency, in which coeffi cients are highly signifi cant at level of 0.1%, so that higher average salary implies higher quality of labour. circumstance is still acceptable. Data of fi rms’ salary are measured in million Vietnamese It is noteworthy that the parameters of exogenous
Table 2. Descriptive statistics for the variables Variable Description Observations Mean Std. Dev. Min Max E Aggregate energy consumption (billion Btu) 1,722 199.74 2557.03 3.4E-04 96885.05 K Total capital (billion VND) 1,722 55.03 130.92 0.137 1629.03 L Total labour (person) 1,722 121.19 158.41 2 2004 Y Added value (billion VND) 1,722 14.73 48.80 -1.84 708.97 gov_sh If having state share in capital (1 = yes, 0 = no) 1,722 0.05 0.22 0 1 type If still using old types of fuel in production (1 = yes, 0 = no) 1,722 0.66 0.47 0 1 size If small or medium-sized (1 = yes, 0 = no) 1,722 0.84 0.37 0 1 p Average price of energy (million VND/million Btu) 1,722 0.72 8.17 6.1E-05 286.483 lql Average salary (million VND/person/year) 1,722 34.75 20.28 1.76471 221.45 im If having import activity (1 = yes, 0 = no) 1,722 0.07 0.25 0 1
PETROVIETNAM - JOURNAL VOL 10/2017 63 PETROLEUM ECONOMICS & MANAGEMENT
Table 3. Results of estimation th in Stata to compute the marginal eff ect of k factor zk Variables Coef. Std. Err. t ratio evaluated at sample mean ( ) is provided
Ln(1/E it ) Frontier equation by Kumbhakar, Wang, and Horncastle [11]. Those results Lnk -0.2 0.06 -3.4*** are shown in Table 4. Lnl 0.92 0.063 14.57*** 5. Discussion LnY -0.92 0.044 -21.07*** Lnk*Lnl 0.01 0.002 3.27** 5.1. Impact of fuel types on energy ineffi ciency Lnk*LnY 0.04 0.007 6.05*** Lnl*LnY -0.03 0.007 -4.2*** The coeffi cient of type on energy ineffi ciency is constant 5.61 0.511 10.97*** positive with signifi cance level of 0.1%, implying higher
u Eit Inefficiency effect equation ineffi ciency in the tile enterprises using conventional fuels gov_sh 0.1 0.041 2.41* (e.g. coal, diesel) than in gas consuming enterprises. type 0.2 0.029 7.1*** More specifi cally, the results in Table 4 indicate size -0.14 0.043 -3.25** lp -0.1 0.013 -7.61*** that the energy ineffi cient gap between fi rms using llql -0.23 0.019 -11.97*** conventional fuels (e.g. coal, diesel) and fi rms using im -0.05 0.047 -1.11 gases is . Thus, we can derive constant 3.44 0.329 10.48*** or . u Variance equation for inefficiency Lne 4.72 0.319 14.77*** This result shows that the level of ineffi ciency in fi rms Lnk 0.69 0.672 1.03 using conventional fuels is higher than that in fi rms using Lnl -4.33 0.755 -5.73*** gases by 1.258 times. In other words, fi rms using gases are LnY 4.12 0.674 6.12*** 1.258 times more energy effi cient than other fi rms. Lnk*Lnl -0.02 0.019 -0.99 It is remarkable that the emission of gas consumption Lnk*LnY -0.2 0.089 -2.24* Lnl*LnY 0.11 0.091 1.22 is much lower than the emission of other fuels. Figure constant -13.58 4.756 -2.86** 2 indicates that natural gas and LPG are the fuels with v Variance equation for error term the least CO2 emission factor. Natural gas and LPG constant -2.16 0.048 -44.97 emit respectively 0.2kg and 0.23kg CO2 for each kWh No. of (equivalent) consumed while other type of fuels, which 1,719 observations are used in the tile production, release from 0.27kg to Note: *: signifi cant at level of 0.05; **: signifi cant at level of 0.01; ***: signifi cant at level of 0.001 0.36kg CO2. Table 4. Average partial eff ect of exogenous variables on According to the database, there are only 40 out
Variables ( k ) (E )/ )Eit ( k of 574 enterprises using LPG or NG in their production. gov_sh 0.27254751 Thus, if the Vietnamese Government would like to type 0.22982667 achieve sustainable development goals, one thing that size -0.30340606 the industry sector and the tile industry, in particular, can lp -0.05659751 help is to encourage as many fi rms to use environmentally friendly energy such as NG and LPG as possible. llql -0.21845733 im -0.05798944 KgCO2/kWh Natural gas 0.2 variables in the ineffi ciency eff ects equation displayed LPG 0.23 0.24 in Table 3 are valueless for economists. The reason is Refinery gas Gasoline 0.25 that they are coeffi cients of pre-truncated distribution Kerosene 0.26 of ineffi ciency u which has truncated distribution, so Eit Diesel 0.27 that they could not explain the real eff ect of exogenous Fuel oil 0.28 variables on uEit. In order to understand the real impacts Hard Coal 0.34 of explanatory factors on energy ineffi ciency, this article Lignite 0.36 calculates the partial eff ect (or marginal eff ect) of each 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 factor via . The formula and the command Figure 2. CO2 emission by type of fuels (kgCO2/kWh) [13]
64 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Of course, the tile producing fi rms may be concerned discussed below), or (ii) many big companies having about the cost of technology conversion and the fact that huge capital are state-related companies (as discussed gas price is higher than the price of conventional fuels. The above), or (iii) the quality of management. The sources of economic effi ciency of technology conversion is not a scope ineffi ciency vary from case to case so it is not possible to of this article and should be considered carefully in another propose a comprehensive solution for all. study. However, based on the data of the tile industry that 5.4. Impact of labour quality on energy ineffi ciency is used in this article, the average cost fi rms using gases need to pay for each million Btu energy consumption is The results in Table 3 indicate that labour quality has about VND 0.412 million, while that number in fi rms using a negative impact on ineffi ciency. Firms that have higher conventional fuels is VND 0.74 million. That comparison quality of labour, which was implied by higher average implies that using gases in production is considerably salary, have lower level of ineffi ciency. In addition, Table cheaper than using other fuels. It is also noteworthy that the 4 shows that for fi rms having higher salary (e.g. 10%) quality and the productivity of tile production using gases will have lower the proportion of energy that needs to are higher than those of production using conventional decrease (e.g. 2.184%). fuels. Therefore, a proportion (or all) of conversion cost and In order to improve energy effi ciency in the energy price diff erence may be absorbed by the spread of Vietnamese tile industry, fi rms should consider improving products’ price. their labour quality, especially with focus on improving 5.2. Impact of ownership structure on energy ineffi ciency the quality of current employees.
The coeffi cients of dummy variable gov_sh are 5.5. Impact of energy price on energy ineffi ciency signifi cantly positive, implying that state-related fi rms The variable lp, as show in Table 3, has a negative (gov_sh = 1) are more ineffi cient than non-state ones eff ect on energy ineffi ciency. However, according to the (gov_sh = 0). marginal impacts recorded in Table 4, that eff ect is pretty From Table 4, we can derive the ratio of input distance small. In detail, in response to 10% increase in energy price, function between a state-related fi rm and a non-state fi rm the energy input distance (or the proportion of consumed as energy needed to reduce) decreases only 0.566%. to prove that the energy input distance of state-related This phenomenon is actually not something fi rms is 1.313 times higher. It implies the ratio of energy unexpected. It is remarkable that the Vietnamese energy consumption to be reduced in state-related fi rms is 31.3% market is not competitive, so that the diff erence in higher than that ratio in non-state fi rms. energy prices here is caused by 2 main factors. Firstly, big contracts made by large fi rms will have more discounts Therefore, in order to increase energy effi ciency, the so that the average cost will be lower than that in smaller tile industry may think about reducing the number of fi rms. Secondly, fi rms using conventional fuels (e.g. coal, state-related fi rms by equitisation or privatisation. diesel…) have smaller average cost than fi rms using eco- 5.3. Impact of enterprise size on energy ineffi ciency friendly fuels such as natural gas and LPG. Because the impacts of the fi rm size and the use of conventional fuels The results from Table 3 show that enterprise size has are already captured in the model, the impact of price on signifi cantly negative impact on energy ineffi ciency in energy ineffi ciency is attenuated. Vietnam. More specifi cally, from Table 4 we can evaluate Nevertheless, energy price is still important and the ratio . It means the should be considered in the ineffi ciency eff ects equation proportion of energy consumption which needs to be because when the Vietnamese energy market becomes cut-off in SMEs is lower than that in large fi rms by 35.4%. more competitive, lp will show more obvious impact on That result is contrary to circumstances in many other energy effi ciency/ineffi ciency. countries where large fi rms have an advantage of scale 5.6. Impact of import activities on energy ineffi ciency to be more effi cient than the others. The reason may be (i) many of the Vietnamese large companies have many Partial eff ect of variable im is considerably small employees but not all of them are productive (tobe in Table 4. Besides, the t ratio of that factor reported in
PETROVIETNAM - JOURNAL VOL 10/2017 65 PETROLEUM ECONOMICS & MANAGEMENT
Table 5. Statistical summary of variables for import activities consumers of other fuels have to pay almost double that Variable Observations Mean Std. Dev. Min Max amount (VND 0.74 million/million Btu on average). Thus, although there may be restrictions about conversion im (if im = 1) 116 1 0 1 1 cost, the Vietnamese tile industry should be encouraged im (if im = 0) 1,606 0 0 0 0 to convert to using environmentally friendly energy to achieve energy effi ciency, less emission and higher Table 3 implies it is not really signifi cant (only signifi cant productivity. at level of 27%). However, since most of materials for the This article has some important contributions for tile industry can be supplied by the domestic market, not stochastic frontier analysis in the energy sector, as well as many companies have import activities. The descriptive for researches on energy ineffi ciency in Vietnam. However, statistic in Table 5 displays that only 116 out of 1,722 certain restrictions still exist in this article. For example, observations have value of im equal to 1 (i.e. fi rm has the scope of time in 3 years is relatively short, or the study import activities). Thus, the signifi cance level of 27% and does not cover some new issues in the SFA such as scaling the appropriate sign of coeffi cient (negative sign) are property of dataset. In addition, there are some potential noticeable results to convince we should not drop it out for further studies as discussed below. of the model. Firstly, this article uses input distance function for It is possible to claim that import activities have a energy ineffi ciency with fi rm-level data. There are several negative impact on the energy ineffi ciency. Therefore, in researches applying the same procedure for province- order to improve energy effi ciency, the Vietnamese tile level data. Thus, other authors in the Vietnamese energy industry may encourage fi rms to be more dynamic in the market can conduct studies with province-level data, import-export market. because energy effi ciency analysis is still really new in 6. Conclusion Vietnam. Secondly, there is a newer model proposed by Green This article implements stochastic frontier analysis [14, 15] which is more fl exible than Battese and Coelli to evaluate energy technical ineffi ciency for fi rm-level [10], but it requires a panel data set for at least 10 years. panel data of the Vietnamese tile industry and the With current statistic data, it is impossible to satisfy that factors aff ecting that ineffi ciency follows the procedure condition. However, in the next few years there will be no of Battese and Coelli [10]. The empirical results fi gure out more constraints and that will be an opportunity to apply the signifi cant impacts of exogenous factors on energy that more advanced model. ineffi ciency, including fuel types, ownership structure, enterprise size, labour quality, energy price and import Furthermore, this article focuses on the tile industry activities. These fi ndings can help the Government due to the limitation in statistic data of other industries have better evidences for considering the strategies to such as steel industry, cement industry. When longer panel achieve the goals of sustainable development. Therein, datasets become available, it is possible to conduct studies the Government may think about some solutions, such on those fi elds and people will have a comprehensive view as fuel conversion, equitisation or privatisation of state- about energy ineffi ciency throughout the industrial sector. related companies, and improving the quality of the Acknowledgements labour force. Concerning the infl uence of fuel types on energy The author would like to thank Professor Chun- effi ciency, this article discovers the diff erence between Hung Kuo in the International University of Japan for his fi rms consuming gases in production and fi rms using continuous support in conducting this research. This work other fuels. More specifi cally, although there are only 40 was supported by the Japanese International Cooperation out of 574 companies in the tile industry that utilise gases, Centre (JICE). the energy effi ciency of these fi rms is 1.258 times better References than that of the remaining fi rms. A comparison of energy costs shows that gas consumers actually need to pay only 1. World Bank. World development indicators 2017. VND 0.412 million/million Btu energy on average, while 2017.
66 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
2. International Energy Agency. Energy balances of 9. Mark M.Pitt, Lung-Fei Lee. The measurement and Non-OECD countries 2017. 2017. sources of technical ineffi ciency in the Indonesian weaving industry. Journal of Development Economics. 1981; 9(1): 3. Gale A.Boyd. Estimating the distribution of plant- p. 43 - 64. level manufacturing energy effi ciency with stochastic frontier regression. US Census Bureau Center for Economic Studies 10. George Edward Battese, Tim J.Coelli. A model Paper No. CES-WP-07-07. 2007. for technical ineffi ciency eff ects in a stochastic frontier production function for panel data. Empirical Economics. 4. Massimo Filippini, Lester Charles Hunt. Energy 1995; 20(2): p. 325 - 332. demand and energy effi ciency in the OECD countries: A stochastic demand frontier approach. Energy Journal. 2011; 11. Subal C.Kumbhakar, Hung-Jen Wang, Alan 32(2): p. 59 - 80. P.Horncastle. A practitioner's guide to stochastic frontier analysis using stata. Cambridge University Press. 2015. 5. Massimo Filippini, Lester Charles Hunt. US residential energy demand and energy effi ciency: A stochastic 12. Martin Painter. The politics of economic restructuring demand frontier approach. Energy Economics. 2012; 34(5): in Vietnam: The case of state-owned enterprise "reform". p. 1484 - 1491. Contemporary Southeast Asia. 2003; 25(1): p. 20 - 43. 6. P.Zhou, B.W.Ang, D.Q.Zhou. Measuring economy- 13. Vietnam Petroleum Institute. Vietnam Petroleum wide energy effi ciency performance: A parametric frontier Institute’s database. 2017. approach. Applied Energy. 2012; 90(1): p. 196 - 200. 14. Willam Greene. Fixed and random eff ects in 7. Satoshi Honma, Jin-Li Hu. A panel data parametric stochastic frontier models. Journal of Productivity Analysis. frontier technique for measuring total-factor energy 2005; 23(1): p. 7 - 32. effi ciency: An application to Japanese regions. Energy. 2014; 15. Willam Greene. Reconsidering heterogeneity 78: p. 732 - 739. in panel data estimators of the stochastic frontier model. 8. Boqiang Lin, Houyin Long. A stochastic frontier Journal of Econometrics. 2005; 126(2): p. 269 - 303. analysis of energy effi ciency of China's chemical industry. Journal of Cleaner Production. 2015; 87: p. 235 - 244.
PETROVIETNAM - JOURNAL VOL 10/2017 67 NEWS
Negotiations on Ca Voi Xanh project to be sped up fi eld is important for ExxonMobil in fostering co-operation with Vietnam in the production and processing of oil and gas products. He gave some proposals to the Deputy Prime Minister on removing obstacles for the parties concerned to early reach the fi nal agreement. Mr. Greenwood also informed that in the near future, ExxonMobil will co-operate with Petrovietnam to research and develop other oil and gas production and processing projects in Vietnam. The Ca Voi Xanh gas fi eld is about 100km east of the central Deputy Prime Minister Trinh Dinh Dung welcomes Mr. Paul Greenwood, ExxonMobil Vice President coast. The Prime Minister has for Gas and Power Marketing. Photo: Nhat Bac approved the plan for construction t a meeting with Mr. Paul Xanh gas fi eld, and stressed that the of 4 gas-fi red power plants with a AGreenwood, ExxonMobil implementation of the project will total capacity of 3,000MW (750MW Vice President for Gas and Power contribute to ensuring the national each power plant) using gas from Marketing on 10 October 2017 in energy security, which is an important Ca Voi Xanh fi eld. Two power plants Hanoi, Deputy Prime Minister Trinh condition for Vietnam to build and will be built in Tam Quang Dinh Dung asked the Ministry of operate gas-fi red power plants in the commune, Nui Thanh district, Industry and Trade, the National central region. Appreciating that the Quang Nam province, and two Oil and Gas Group (Petrovietnam), project has achieved some important power plants will be built in Dung and ExxonMobil to accelerate progress lately, he reaffi rmed Quat economic zone in Binh Thanh negotiations on the Ca Voi Xanh the Vietnamese Government’s commune, Binh Son district, Quang project on the basis of ensuring the commitment to creating favourable Ngai province. The Prime Minister harmony of interests of the parties. conditions for the fi rst gas to be has also agreed to use part of the received from Ca Voi Xanh fi eld in 2023. gas recovered from Ca Voi Xanh The Deputy Prime Minister lauded fi eld to develop the petrochemical ExxonMobil for its engagement in Mr. Greenwood said the gas industry. developing and exploiting the Ca Voi recovery project in Ca Voi Xanh Nguyen Hoang
Nhon Trach 2 Power Plant reaches production milestone of 30 billion kWh hon Trach 2 Power Plant (PVPower NNT2) reached production milestone of 30 billion kWh of electricity supplied to the national grid at 22:15 hours on 26 October 2017. This is a great eff ort of PVPower NT2’s staff and engineers after 6 years of putting PVPower NT2 into commercial operation. Meanwhile, other power plants with similar capacity (750MW) need 7 - 8 years of commercial operation to achieve 30 billion kWh. Thuy Hang Nhon Trach 2 Power Plant. Photo: PVN
68 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
PVN co-operates with potential partners to manufacture polyester fiber partner’s specialists, the plant has modern technology and equipment which have been preserved in good conditions and its operation can be restarted within a short time, 2 months at the latest. Based on the working results between the two parties, Petrovietnam and the partner agreed to sign a comprehensive co- operation agreement with potential domestic and foreign partners in the oil refi ning, petrochemical and polyester fi ber sectors, including support for PVTEX in the preparation and restart of the plant. Dinh Vu Polyester Fiber Plant. Photo: PVN After signing the agreement, he Vietnam Oil and Gas Textile Fiber JSC (PVTEX), the investor the partners will continue with TGroup (Petrovietnam) of Dinh Vu Polyester Fiber Plant, technical and cost assessments, and signed a comprehensive co- worked with domestic and foreign work with the domestic partners operation agreement with potential partners to fi nd solutions to remove to propose a detailed co-operation domestic and foreign partners in diffi culties and restart the plant. plan. Accordingly, the partners will the oil refi ning, petrochemical, and provide comprehensive support In March 2017, Petrovietnam and polyester fi ber sectors on 5 October to PVTEX in quality management, PVTEX worked with a foreign partner, 2017 in Hanoi. technical staff and operational one of the leading corporations worker training, supply of materials Following the guidance of the operating in the fi eld of polyester fi ber with stable quality at market prices, Government and the Ministry of (its total fi ber production reached 3.5 reduction of costs, and improvement Industry and Trade, Petrovietnam million tons in 2016). According to of the plant’s operation effi ciency. and Petrovietnam Petrochemical and the preliminary assessment by the Hong Ngoc
Song Hau 1 Thermal Power Plant: Main transformer for Machine Unit 2 installed n 27 October 2017, the Song Hau 1 Thermal Power Plant OManagement Board of project consists of two machine units Song Hau 1 Thermal Power Plant with the total capacity of 1,200MW project, the EPC contractor - Vietnam (2 x 600MW), construction area of Machinery Installation Corporation 115.2ha, invested by the Vietnam Oil (Lilama) in co-ordination with and Gas Group. The total investment Doosan and Hyundai carried out the of the project is about VND 43,043 installation of the main transformer of billion, equal to USD 2,046 billion. Machine Unit 2 in Song Hau 1 Thermal Scheduled for coming online in Power Plant. The main transformer, 2020, Song Hau 1 Thermal Power manufactured by Hyundai (Korea), Plant will supply 7.8 billion kWh/year has the weight of about 560tons, the for the national grid, contributing voltage level of 525/22kV and the to assurance of the national energy capacity of 730MVA. security. Thuy Hang Song Hau 1 Thermal Power Plant. Photo: PVN
PETROVIETNAM - JOURNAL VOL 10/2017 69 NEWS
PVN and PVEP sign outsourcing contract for operation of Blocks 01 & 02 Since the handover, PVEP has established a project management board for Blocks 01 and 02 to ensure safe, continuous and most productive operation and management of the project. One month after the handover, PVEP has managed and operated Blocks 01 & 02 to produce an output of roughly 360 thousand barrels of oil, earning a total revenue of roughly USD 20.52 million (at the Dr. Nguyen Quynh Lam, Vice President of Petrovietnam and Dr. Ngo Huu Hai, President & CEO price of USD 57.5/barrel). of PVEP sign the contract. Photo: PVN n 9 October 2017, the Petronas Carigali Vietnam Limited President & CEO of PVEP Ngo OVietnam Oil and Gas Group (PCVL) and PVEP. This is among the Huu Hai pledged that PVEP will pro- (Petrovietnam) signed an outsourcing fi rst generation of Petrovietnam's actively and creatively implement contract for operation of Blocks 01 oil and gas contracts and the fi rst necessary measures to maintain the production output, improve oil and 02 with Petrovietnam Exploration to conclude upon maturity on 9 recovery rates, and ensure the safest Production Corporation (PVEP). September 2017 after 26 years of operation. Petrovietnam then signed operation of the wells in Blocks 01 The Production Sharing Contract an agreement to handover the assets and 02 to bring about the highest (PSC) for Blocks 01 and 02 was signed and operations of Blocks 01 and 02 to effi ciency for the Government and on 9 September 1991 between PVEP. Petrovietnam. Manh Hoa
VPI signs co-operation agreement with Industrial University of Tyumen Also at the Tyumen Oil and Gas Forum 2017, Dr. Nguyen Hong Minh attended the discussion on human resource training and development and presented a report on Petrovietnam’s concepts and practical implementation of oil and gas specialist training and development activities. During the mission in Russia, the VPI delegation also worked with experts and some units under IUT
Acting Rector of the Industrial University of Tyumen Veronika Efremova and Deputy General Director of Vietnam Petroleum on advanced technologies for oil Institute Nguyen Hong Minh sign a co-operation agreement. Photo: oilandgasforum.ru spill treatment and enhanced oil he Vietnam Petroleum of training high quality specialists recovery. In the immediate future, VPI will invite some Russian experts TInstitute (VPI) has signed a (master's and doctor’s degrees); to teach and impart knowledge co-operation agreement with the organising internship programmes and experience on oil and gas Industrial University of Tyumen (IUT) for students, exchanges of offi cials exploration and production in within the framework of the Tyumen and specialists; conducting joint the Russian Federation for the Oil and Gas Forum (Russia). research projects and research- Vietnamese staff . Accordingly, VPI and IUT will training programmes in the fi elds of Chi Linh strengthen co-operation in the areas oil and gas and related industries.
70 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Amended gas sales and purchase and transportation contracts signed for Block 06-1, Nam Con Son basin n 16 October 2017, a Osigning ceremony was held at Petrovietnam’s Headquarters for the 3th amended and supplemented contract of Block 06-1’s gas sales and purchase contract between the Vietnam Oil and Gas Group (Petrovietnam), Rosneft Vietnam B.V., ONGC Videsh Ltd. and Petrovietnam Gas Joint Stock Corporation (PV GAS); and the 3th amended and supplemented contract of Block 06-1’s gas and condensate transportation, treatment and management contract between Signing Block 06-1’s amended gas sales and purchase contract. Photo: Nhu Trang Petrovietnam, Rosneft Vietnam B.V., ONGC Videsh Ltd. and the members of gas per day and 1,500 barrels of operation to expand prospecting of Nam Con Son Gas Pipelines condensate per day. and exploration activities in the Company (including PV GAS, Perenco coming period. Petrovietnam’s President and CEO Vietnam A.S. and Rosneft Pipeline Nguyen Vu Truong Son confi rmed Phong Lan Dai fi eld was Vietnam B.V.). that the signed contracts facilitate discovered in 2016, and its Outline Block 06-1 project (Nam Con Son the development of Phong Lan Development Plan (ODP) was basin), developed by the consortium Dai fi eld and supplementary approved by the Ministry of Industry of contractors including ONGC development of Lan Do fi eld to and Trade on 3 January 2017 with the target of producing the fi rst gas Videsh Ltd. (45%), Rosneft Vietnam optimise the equipment and assets B.V. (35%) and Petrovietnam (20%), in the fourth quarter of 2018. Gas of Block 06-1 and ensure the stable is producing gas and condensate in and condensate from Phong Lan Dai production output of Block 06-1 in Lan Tay, Lan Do and Phong Lan Dai fi eld will be transported through the the future. fi elds. Block 06-1 has been operating Nam Con Son pipelines to Dinh Co stably for the last 15 years with an Contractors participating in Gas Processing Plant and Phu My Gas- 3 average production of 8.8 million m Block 06-1 will strengthen co- Power-Fertilizer Complex. Hong Ngoc PVEP strengthens co-operation with Russian petroleum partners n 2 October 2017, Dr. Ngo Huu Hai, OPresident and CEO of the Petrovietnam Exploration Production Corporation (PVEP), and his delegation worked with Mr. Sergey Tumanov, CEO of Gazprom EP International in Saint Petersburg. The two sides discussed the production and business situation, the results of implementing works at oil and gas blocks in Vietnam and the prospects for future co-operation. During the visit to Russia, Dr. Ngo Huu Hai attended the 50th Anniversary of Zarubezhneft’s establishment in Moscow. Manh Hoa Dr. Ngo Huu Hai attends the 50th Anniversary of Zarubezhneft’s establishment. Photo: PVEP
PETROVIETNAM - JOURNAL VOL 10/2017 71 NEWS
Proposed measures to boost production in Bir Seba project, Algeria as well as investment effi ciency for PVEP and the partners. At the meeting with Alnaft President Arezki Hocini, the Petrovietnam/PVEP delegation conveyed the message of the Bir Seba Project’s investors and proposed specifi c measures to support the project. The Bir Seba fi eld development project in Blocks 433a and 416b, Sahara desert, Algeria has been Bir Seba fi eld. Photo: PVEP operated by Groupment Bir Seba (GBRS) with the participating rom 14 - 17 October 2017, Seba fi eld development project in interests of PVEP (40%), Sonatrach FPetrovietnam Vice President Blocks 433a & 416b, Algeria. (25%) and PTTEP (35%). So far, Nguyen Quynh Lam worked with At the meeting with CEO of accounted for PVEP only, the project’s Algeria’s national oil and gas Sonatrach Ould Kaddour, the recoverable reserve belonging to company (Sonatrach) and Algerian Petrovietnam/PVEP leaders have PVEP alone amounts to 11 million National Agency for the Valorisation proposed measures and mechanisms tons of oil and its production output of Hydrocarbon Resources (ALNAFT) to improve and enhance the is almost 20,000 barrels per day. on the implementation of the Bir production capacity of the project Nguyen Hoa
Phu My Fertilizer Plant meets target 2 months ahead of schedule n 1 November 2017, Phu My OFertilizer Plant’s production reached 777 thousand tons of urea equivalent, equal to 101% of its plan for 2017. This is the 11th consecutive year in which the plant completed its production plan ahead of time. The Petrovietnam Fertilizer and Chemicals Corporation (PVFCCo) informed that it is preparing for the periodical overall maintenance of Phu My Fertilizer Plant by the end of November 2017. This is the largest and most complex overall maintenance ever to date with over 4,000 work items because the tie- in project to increase the capacity of NH unit will be simultaneously 3 Phu My Fertilizer Plant. Photo: PVFCCo implemented during this time. 90% of the 2017 plan); 330 thousand plan). Total consolidated revenue is By the end of October 2017, tons of other fertilizers (equal to estimated at VND 7,078 billion and PVFCCo has sold about 700,000 tons 103% of the plan) and 27,000 tons pre-tax profi t is estimated at VND of Phu My Fertilizer (equal to nearly of chemicals (equal to 129% of the 800.2 billion. Bui Ha
72 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
KVT meets LPG and Bach Ho condensate production targets ahead of time system at Dinh Co Gas Processing Plant reaches 100%; the level of readiness of the gas processing and supply system is constantly at 100%; the readiness of equipment aff ecting liquid product recovery is 99.948%. At the same time, KVT has implemented measures to optimise operation and maintenance such as early and continuous evaluation of the infl uence of gas components from new fi elds like Dai Hung and Dinh Co Gas Processing Plant. Photo: PV GAS Thien Ung to develop optimal etrovietnam Gas Vung Tau year of 2017, KVT will produce 285 operation plans and parameters; PProcessing Company (KVT), thousand tons of LPG, equivalent test and apply solution to fl oods a member of Petrovietnam Gas to 124% of the 2017 plan and Bach at the top of tower C-05; develop Joint Stock Corporation (PV GAS), Ho condensate output will reach 72 plans to shorten the maintenance has completed its 2017 annual thousand tons, equivalent to 136% time of critical equipment aff ecting production targets ahead of time. of the 2017 plan. the liquefi ed product recovery performance. The company has By 14 October 2017 inclusive, LPG To achieve these results, the co-ordinated with DVK and DAK to output reached 232 thousand tons company has taken measures to develop the maintenance and repair (equal to 100.4% of its plan for 2017 recover as much liquefi ed products plan during the non-productive and 79 days earlier than schedule); as possible such as ensuring safe and time in 2017 at GPP and tie-in the Bach Ho condensate output effi cient operation of gas facilities, enhanced LPG recovery project to amounted to 58.1 thousand tons maintaining a high level of readiness ensure safety, quality and progress (equal to 109.73% of its plan for 2017 and reliability of its gas processing so as to increase the recovery of dry and 102 days earlier than schedule). system. The results showed that gas and liquid products. Minh Ngoc It is expected that for the whole the reliability of the gas processing
Ca Mau Gas Processing Plant supplies first LPG shipment fter nearly one month of Atrial operation, Ca Mau Gas Processing Plant has completed the technical and equipment inspection, and supplied the fi rst commercial LPG shipment for tank trucks on 5 October 2017. Ca Mau Gas Processing Plant with the capacity of 6.2 million m3 of gas/day is a crucial part of the Southwestern gas supply chain, recovering LPG and condensate from natural gas sources to contribute to Ca Mau Gas Processing Plant. Photo: PVN enhancing gas economic value. With time, the storage area and port of and stabilising the price of LPG in the 4 LPG tanks having a total capacity of Ca Mau Gas Processing Plant play an South West region. Viet Hung 8,000 tons and 30 years of operating important role in storing, supplying
PETROVIETNAM - JOURNAL VOL 10/2017 73 OIL & GAS MARKET
WORLD OIL AND GAS MARKET
or the fi rst time in more than 80 F2 years, Brent crude oil price for December delivery, rising above 70 USD 60/barrel, moved toward USD 61/barrel on 30 October. While the 60 US light, sweet (WTI) December crude price settled above USD 54/ USD/Barrel 50 barrel - its highest settlement since 40 February. Most analysts agreed Brent WTI oil prices rose on recent optimism 30 that the Organization of Petroleum
Exporting Countries (OPEC) and other 01/09 03/09 05/09 07/09 09/09 11/09 13/09 15/09 17/09 19/09 21/09 23/09 25/09 27/09 29/09 01/10 03/10 05/10 07/10 09/10 11/10 13/10 15/10 17/10 19/10 21/10 23/10 25/10 27/10 29/10 major producers likely will extend Figure 1. Brent and WTI spot prices from September to October. Source: EIA production-cut targets beyond March especially in China and India, the industry sources and Thomson 2018 to help rebalance oil supply- world’s number one and three Reuters Analytics showed. demand levels on world markets. importers. China’s oil demand Given the tightening oil market remains voracious, hitting a January According to William O’Loughlin, conditions, many analysts expect to September average of 8.5 million investment analyst at Rivkin prices to rise further. Shane Chanel, barrels/day. Three main factors were Securities, oil prices are holding equities and derivatives adviser at driving China’s insatiable appetite comfortably above USD 50/barrel ASR Wealth Advisers expect oil prices for crude: declining domestic as possible supply disruptions in higher by 10% by the end of the production, increased access to the Kurdish region of Iraq support year and have started to accumulate imports and exports for independent prices. US production was also strong positions within the oil sector. refi ners, and building up the strategic recently impacted by a hurricane for petroleum reserve, Britain’s Barclays the second time in as many months IEA’s October 2017 oil market bank said. Besides, India’s fuel thirst and the number of US drilling rigs report is also increasing. India imported a declined for the third week in a row. record 4.83 million barrels/day of In Oil Market Report for October, In Baker Hughes report, the US rig oil in September as several refi ners IEA noted that falling global crude count during the week ended 27 resumed operations after extensive oil stockpiles in 2017 will help put October declined for the tenth time maintenance to meet rising local fuel the market “roughly” into balance in in 13 weeks. Baker Hughes’ tally of demand. The country’s September 2018, but an increase in prices could active rigs fell to 909. imports stood 4.2% above this time be limited, especially if OPEC doesn’t In the main growth areas of last year and about 19% more than stick to its agreement to curb output. Asia, consumption remains strong, in August, ship-tracking data from The recent upward momentum
74 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Venezuela and Algeria, plus non-OPEC Russia and Oman, were meeting in Vienna after oil prices gained more than 15% in the past three months to trade above USD 56/barrel. Russia’s budget revenue from oil and gas taxes, which earlier this year recovered to 2015 levels, fell again by July on crude price fl uctuations and a stronger ruble. Russian economic growth accelerated to the fastest pace in almost fi ve years in the second quarter of 2017 amid recovery in oil prices and domestic consumption. The nation has benefi ted from the crude-output pact with OPEC and other producers and could benefi t Figure 2. Global demand - supply balance prediction until 2018. Source: IEA’s October 2017 Oil Market Report from extending the accord. in crude prices was provided by Based on unchanged OPEC It may not be possible to discuss uncertainty with suppliers such as output and normal weather such an extension earlier than Libya, Venezuela, Iran, and Northern conditions, IEA expects three of January 2018 but it should be phased Iraq, signs of possibly slower- four quarters in 2018 to be “roughly out in a way that ensures all positive than-expected growth in US shale balanced,” with a fi rst-quarter stock results of the pact are retained once production, and strong oil demand. build of 800,000 barrels/day. the deal ends, according to Russian Meanwhile, OPEC crude output was Energy Minister Alexander Novak in But IEA projects oil demand and an interview after the meeting. He virtually unchanged in September non-OPEC production in 2018 will sees oil at USD 50 - 60/barrel through at 32.65 million barrels/day, down grow by about the same volume, the end of 2017, which would be a 400,000 barrels/day year-over-year, which could act as a ceiling for oil good price both for producers and as slightly higher supply from Libya prices. Non-OPEC output is expected suppliers. Based on this argument, and Iraq off set lower supply from to increase 700,000 barrels/day in OPEC and Russia said they were about Venezuela. Year-to-date compliance 2017 and 1.5 million barrels/day halfway toward clearing a global oil with the group’s agreement to in 2018. Projected global demand glut and urged fellow producers to curtail output by 1.2 million barrels/ growth remains at 1.6 million barrels/ stay focused and fi nish the job, while day is 86%. day in 2017 and 1.4 million barrels/ stopping short of additional action day in 2018. In the Organization for Economic to reassure a jittery market. At the Cooperation and Development The extension ability of the agree- meeting in Vienna mentioned above, (OECD), the 5-year average stock ment to cut production signed in ministers hailed the accelerating overhang is down to 170 million 2016 market recovery as Brent crude, the barrels from 318 million barrels at international benchmark, climbed to the end of January. Stocks have On 22 September 2017, in Vienna, a six-month high. Still, predictions of fallen in months when they normally OPEC, Russia and several other strong growth in US shale production increase, off setting net builds in producers had held a meeting to next year means there are still China, where crude imports have review the accord to cut production concerns that the scheduled expiry fallen every month since June 2017, by about 1.8 million barrels/day since of the agreement at the end of March and the implied net build for stocks January 2017 that expires in March could be premature. Thus, the Joint in September was relatively small at 2018. Ministers on a panel monitoring Ministerial Monitoring Committee 100,000 barrels/day. the pact, comprising Kuwait, would consider extending the
PETROVIETNAM - JOURNAL VOL 10/2017 75 OIL & GAS MARKET
Table 1. World oil supply and demand (million barrels/day)
2017 2018 1Q 2Q 3Q 4Q Avg. 1Q 2Q 3Q 4Q Avg. DEMAND1 96.6 97.8 97.9 98.5 97.7 97.9 99.1 99.5 99.9 99.1 OECD 46.9 47.0 47.4 47.7 47.3 46.9 46.9 47.7 47.7 47.3 Americas 24.5 25.0 25.0 25.2 24.9 24.6 25.1 25.3 25.1 25.0 Europe 13.9 14.2 14.6 14.3 14.3 13.8 14.3 14.7 14.4 14.3 Asia Oceania 8.6 7.8 7.8 8.3 8.1 8.4 7.6 7.7 8.2 8.0 NON-OECD 49.7 50.8 50.5 50.8 50.4 51.0 52.2 51.8 52.2 51.8 FSU 4.6 4.8 5.0 4.9 4.8 4.7 4.8 5.1 5.0 4.9 Europe 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.8 0.7 China 12.5 12.7 12.1 12.3 12.4 12.8 12.9 12.5 12.7 12.7 Other Asia 13.2 13.4 13.0 13.7 13.3 13.8 14.0 13.6 14.2 13.9 Americas 6.4 6.6 6.7 6.6 6.6 6.5 6.7 6.8 6.7 6.7 Middle East 7.9 8.4 8.8 8.3 8.4 8.1 8.6 8.9 8.4 8.5 Africa 4.4 4.2 4.1 4.2 4.2 4.4 4.3 4.2 4.4 4.3 SUPPLY 96.6 97.8 97.9 98.5 97.7 97.9 99.1 99.5 99.9 99.1 OECD 24.0 23.7 23.8 24.4 24.0 25.1 25.0 25.2 25.8 25.3 Americas4 20.0 19.8 19.9 20.4 20.0 21.0 20.9 21.3 21.7 21.2 Europe 3.7 3.5 3.4 3.6 3.6 3.7 3.7 3.5 3.6 3.6 Asia Oceania 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.5 0.4 NON-OECD 57.7 57.8 58.1 58.6 58.1 59.0 59.4 59.8 60.2 59.6 FSU 14.5 14.4 14.3 14.4 14.4 14.5 14.5 14.4 14.4 14.5 Europe 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 China 3.9 3.9 3.8 3.8 3.9 3.8 3.7 3.7 3.7 3.7 Other Asia2 3.5 3.5 3.5 3.4 3.5 3.4 3.4 3.3 3.3 3.4 Americas2,4 4.6 4.5 4.6 4.7 4.6 4.6 4.7 4.8 4.9 4.8 Middle East 1.2 1.2 1.2 1.2 1.2 1.2 1.2 1.3 1.3 1.2 Africa2 1.7 1.7 1.8 1.7 1.7 1.8 1.8 1.8 1.8 1.8 Total Non-OECD 29.5 29.3 29.3 29.4 29.4 29.5 29.5 29.4 29.5 29.5 Processing gains3 2.3 2.3 2.3 2.3 2.3 2.3 2.3 2.3 2.3 2.3 Global Biofuels 1.9 2.5 2.8 2.5 2.4 2.1 2.6 3.0 2.6 2.5 OPEC 38.9 39.2 39.7 Crude 32.1 32.3 32.7 NGLs 6.9 6.9 6.9 6.9 6.9 7.0 7.1 7.0 7.0 7.0 1Measured as deliveries from refi neries and primary stocks,comprises inland deliveries, international marine bunkers, refi neryfuel, crude for direct burning, oil from non-conventional sources and other sources of supply. Includes Biofuels. 2Other Asia includes Indonesia throughout. Latin America excludes Ecuador throughout. Africa excludes Angola, Gabon and Equatorial Guinea throughout. 3Net volumetric gains and losses in the refi ning process and marine transportation losses.4 Comprises crude oil, condensates, NGLs, oil from non-conventional sources and other sources of supply. Source: IEA’s October Oil Market Report supply cut pact and can make The impact of hurricanes outages, the processing of oil policy recommendations for the into fuels slows down. This leads According to Steven Austin’s wider group of OPEC and non-OPEC to turbulent oil prices because of article published on oil-price.net, the producers, which meets in 2017. Harvey hurricane caused refi nery short supply. But at the same time, According to IEA, with OPEC shutdowns, fl ooded wells, port hurricanes and storms results in members scheduled to meet in closures and oil production outages. lower gasoline demand due to lower Vienna on 30 November, the next Within the various oil sub-sectors, the consumption. So, in truth, the eff ects few weeks will be crucial in shaping consequences of Harvey are more dire of a storm are felt only after a week their decision on output. A lot has for downstream, especially refi ning. of the real event, when downstream been achieved towards stabilising the When refi ning capacity is operations return to normalcy. market, but to build on this success in aff ected, due to natural events Hurricanes Harvey, Irma, and Jose… 2018 will require continued discipline. like hurricanes or due to technical were no exceptions.
76 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Source: Dgcglobal.com
The South-Eastern Texas coast, the storm has cut off about 24% of were on lockdown because of where Harvey stuck, is a major oil oil production in the Gulf of Mexico Harvey. Clearing the debris from the hub fi lled with ports, terminals and and 26% of natural gas production. channels and ports so as to enable refi neries. Particularly, Houston, Furthermore, the Gulf Coast the smooth fl ow of crude oil to Texas city and Baytown regions contributes almost one-fi fth of total refi neries will take some time. oil output in the US. It also houses have eleven refi neries that together About the pipelines, the main about one-third of US’s capacity to manage 2.7 million barrels/day, diesel, jet and gasoline lines of turn oil into gas, diesel and other by or about 14% of the US refi ning Colonial Pipeline - the biggest fuel products. Harvey knocked off about capacity. In fact, more than 3 million system in the US that transports a quarter of that oil, nearly 430,000 barrels of crude move to the Gulf more than 100 million gallons of barrels/day, produced in the Gulf of Coast everyday from countries like heating oil, gasoline, and aviation Mexico. Mexico, Saudi Arabia, Venezuela and fuel from Houston, have been Columbia. About 1 million barrels of To be sure, more than 20% of US temporarily shut because of lack of crude are also exported each day. refi ning capacity have been aff ected. fuel. Along the same lines, the main According to the US Bureau of Safety According to the Energy Department line of the Explorer Pipeline with and Environmental Enforcement, about 10 refi neries in the Gulf Coast a capacity of 660,000 barrels/day
PETROVIETNAM - JOURNAL VOL 10/2017 77 OIL & GAS MARKET
hauling fuel from the Gulf coast to Harvey, that outlook would remain the Midwest region has been shut a dream for some time. On the other too. Two East refi neries have run out hand, the oil drillers have returned of gasoline meant for immediate to work in the Gulf coast. As far as delivery. More than 1 million barrels/ the refi neries are concerned, the day of oil is expected to remain aftermath has more to do with offl ine in the Houston/Galveston fl ooding rainfall than mechanical area alone. In total, Harvey shunted damages. So, the refi neries are about 4.4 million barrels of oil per expected back in operation soon. day of refi ning capacity. Motiva refi nery, the US’s largest with 600,000 barrels/day capacity, Meantime, the US Government had already restarted about 40% has approved the release of about production within days. Likewise, 4.5 million barrels of crude from port operations are promptly the Strategic Petroleum Reserve resuming along the US Gulf coast. (SPR). Earlier, the withdrawal of 1 million barrels from the emergency At present, Europe, with a stockpile was announced for a surplus of gasoline, has stepped in refi nery in Lake Charles, Louisiana with additional gasoline shipments. and 500,000 barrels to Valero However, as soon as the refi neries Energy. The withdrawals, appearing return to normalcy, the shipments signifi cant while making news, from Europe will fall back. are more of a politically motivated Hurricanes are a yearly symbolic gesture. SPR does protect occurrence. Unlike earthquakes, the domestic economy from outside hurricanes are predicted weeks market threats like the OPEC cartel. in advance allowing remediation However, it is woefully useless in the measures to be put in place and facts of hurricanes because, simply minimise impact. The Texan put, if refi neries are down, more oil downstream sector was up an isn’t going to help. running within days of Harvey between Saudi Arabia, US, Russia, Meanwhile, OPEC and other oil hurricane. Iran and the oil market share war. As exporting countries are striving to The downside of low oil prices a result, the whole market dynamics keep off 1.8 million barrels/day of have deeply changed. Crude oil is no crude off the market to lower the At fi rst sight, it appears that low longer a commodity whose price is crude oil stockpiles of oil importing oil prices are good for society. A low controlled by a cartel. It has become countries as discussed above. But oil price helps us save money at the a commodity subject to supply and the weaker demand will keep the pump, acts as an economic catalyst demand like any other commodity. calculations continually wrong. In making cheaper exchanges and And since there is plenty of supply, the US, the crude oil inventories thereby stimulating the economy. crude oil prices remain fl at for a long stand at 463.2 million barrels. Crude However, for the oil industry and the time. This new dynamics means that oil imports in the US averaged 8.8 oil net export countries, especially oil prices will remain about USD 55 - million barrels/day with an increase the downside of low oil prices creates 60/barrel between whiles. of 664,000 barrels/day, according to more complicated problems that the EIA. have addressed today. Money leaves long-term projects
Not long ago, the oil industry For the fi rst time, we have Because of oil price volatility, was banking on a sunny target of 10 entered an era of sustained low oil the investment world is shying from million barrels/day in oil production prices caused by the global economic large funds needed in ambitious by the end of 2017. Because of crisis combined with the oil price war projects in favour of more modest
78 PETROVIETNAM - JOURNAL VOL 10/2017 PETROVIETNAM
Stena Carron drill ship. Source: Exxon Mobil fracked wells with a near-immediate 2.25 million barrels/day in 2020, investors, but also Middle Eastern production and favourable return which will put the US alongside and Saudi investors. on investment (ROI). Fracking the Kuwait and UAE. That is four times For good measure, US has technology whereby oil is extracted higher than 2016 levels. Currently, strengthened the infrastructure from shale has changed the energy US exports are about 500,000 with regards to energy projects. landscape of the US. It has enabled barrels/day of oil and rising. And the It seems another age when crude oil production to ramp up by 10% proceeds are usually reinvested or oil export was banned in the US. since mid-2016 to 9.34 million paid out as dividends to investors - Drifting across the pipelines, shale barrels/day. Astonishing considering in any cases turned into something oil has brought a seismic shift with that fracking contributed less than productive, not taxed out to sustain regards to how we view energy 2% to US oil production in 2000. With Middle Eastern welfare theocracies. in the US. Shale explains the the road still long and unchartered, Therefore, the economic model of increased oil production. And this this is just the tip of the iceberg in the Gulf conventional oil resources seems is a chance for investors. After all, history of shale. a tad riskier. It makes economic sense smaller incremental investments for investors to divert funds from With strong basics, all signs point help hedge risks. The fl ood of new Middle-East exploration to US shores to the assumption that oil production money into projects with immediate and other regions with an eye on will continue to surge in the US. US results has resulted in a renewed the-better-long term outlook. In fact, crude exports are estimated to touch oversupply of fresh oil from the US. it includes not only US and Western
PETROVIETNAM - JOURNAL VOL 10/2017 79 OIL & GAS MARKET
Valhall platform. Source: norskpetroleum.no
Investors shy Gulf State oil projects Investors are dubious a lot of Back to the question of how things when it comes to the IPO. Middle-East can tide the loss, it’s Amin Nasser, CEO of Saudi Many analysts doubt whether the about going back to the basics. Now Aramco, said that the lack of IPO is worth the touted USD 2 trillion. that cash is hard on the pocket, the investments in conventional oil fi elds Revenue from Aramco contributes industry has to rethink, grasp and opt was caused by unconventional oil to the operating budget of the for better planning and technology and gas in the market reeks of the Kingdom so investors would be to increase effi ciency in each project. pot calling the kettle black. These wary of the political ramifi cations. statements are not helping Saudi Obviously, it’s fi nancing that Aramco’s claim to 260 billion barrels Aramco’s IPO set for 2018, which gets oil out of the ground. When of recoverable oil being one of them. could use more help to lure investors. investment dries up, the pump But that’s not the biggest reason: In fact the IPO of this trillion-dollar getting oil out of the ground gets Saudi investors are also leaving Saudi company touted as “the biggest in put away. With sluggish oil prices Arabia to invest in the US petroleum history” could not have come at a what we have in hand are wary and industry instead. This really gives worst time given how the Kingdom over-cautious speculators. As it is Saudi Aramco the allure of a sinking has in short order managed to lose right now, the big money is betting ship. control of OPEC. To cut the story on US fracking instead conventional short, it was the master plan of Saudi The world’s biggest crude exploration in Gulf States, and that Arabia that forced OPEC to overreact exporter is conceding ground to includes Middle Eastern investors. in cutting oil production to boost oil enact a steep oil production cut in Since Gulf projects will not be prices that eventually culminated in order to prop up the low global price renewed at a sustainable rate, these Saudi Arabia losing market share and of oil, causing the Kingdom to rapidly are long terms changes that are clout in OPEC. lose market share and infl uence over reshaping the industry for good. other OPEC members. Collected by Quang Trung - Ngoc Toan
80 PETROVIETNAM - JOURNAL VOL 10/2017