143 NZOH Volume 11 Number 2 July 1983

COAL STRATEGIES FOR THE HUNTLY

W. R, BAKER

UNIVERSITY OF CANTERBURY CHRISTCHURCH,

SU1MARY

The development and use of a method (1) for determining least cost strategies at the Huntly power station is described. The core is a simulation model of the New Zealand power system with two regions, load/flow variation within the year, stochastic hydro inflows and reservoir management rules. The latter are developed by stochastic dynamic programming. The method treats uncertainties in power requirements, oil prices and coal output build-up via scenarios. The method enables explicit consideration of possible regret in optimisation.

1. I N T R O D U C T I O N

Huntly Power Station (HPS) is a 960 MW currently being commissioned in the of New Zealand. The station can use either or local coal for fuel. Gas availability is limited and specified. So the HPS considered is in reality the coal fuelled section of the station, which is fairly low in the merit order of power station fuel costs. Therefore HPS will probably initially provide energy firming by generating at high loads only when the dominant hydro generation system experiences low inflows. Average annual HPS generation should increase as electricity demands increase, with the station being baseloaded eventually. HPS's energy firming fuel use may vary greatly within and between years. When base- loaded at 70% plant factor HPS requires about 2,600,000 tonnes of coal to generate 5800 gWh/yr.

Waikato coal production must increase greatly to meet forecast new demands for coal, principally for and steel production. The recently commissioned Huntly West under­ ground mine has a design maximum output of 1,200,000 tonnes per year and is to be the major supplier of HPS. Additional new underground and/or opencast production will eventually be required. Underground production rates are inflexible for technical and man­ power reasons . A stockpile is therefore required to absorb dif­ ferences between supply of coal to HPS and use of coal at HPS, particularly under energy firming conditions. Long planning and construction lead times, and the associated high regret costs possible, motivate this search for optimal coal supply and stock­ piling strategies.

Manuscript submitted January 1983 144

2. METHODOLOGY FOR OPTIMISATION

Decisions must be made as to the stockpile capacity, Huntly West buildup rate, and timing and quantity of new coal production for HPS. Decisions should further the generation system aim of least cost electricity supply. Uncertainty surrounds future hydro inflows, electricity demand, oil prices, and generating plant additions. The value of having coal for generation varies greatly within years as a reflection of annual electricity demand cycles and hydro inflow cycles. Overburden stripping is dependant on dry (summer) conditions. An outright optimisation model should therefore be stochastic in several dimensions and cover subannual variations. Currently such models are insoluble because of com­ putational requirements, so a less complex formulation is devel­ o ped as follows.

The number of possible sets of coal decisions (hence termed 'policies') is limited by applying a discrete approximation to each continuous decision variable. Quarterly time periods and a fifteen year horizon are adopted. Scenarios are developed cover­ ing possible future outcomes of the 'behavioural' uncertainties; i.e. electricity demand, oil prices and plant additions. The scenario elements are outlined in Table 1. For each scenario hydro inflow distributions and hydro storage operating rules are developed for each period. A simulation model of the power system, with detailed coal operations, is used to determine the expected cost over hydro inflow possibilities of a specified coal policy and scenario. For each scenario the policy is success­ ively altered, until no further cost reduction can be found, to produce a 'best' policy. The set of best policies guides identi­ fication of an overall optimum strategy which fixes only current d e c i s i o n s .

Table 1: Scenarios Formed

NAME 4 % 4 D E M A N D OIL PRICE HYDRO THERMAL SECOND PLANT PLANT SMELTER

PAUCITY1 Yes High High Delay NZE Yes HIDD Yes High Central NZE N Z E Yes H D E L Y No C e n t r a l C e n t r a l Delay NZE Yes T D E L Y No C e n t r a l C e n t r a l NZE Delay Yes HIOIL No Central High NZE NZE Yes C E N T R A L 2 Yes Central Central N Z E N ZE Yes L O O I L No C e n t r a l L o w NZENZE Yes NOSM No Central Central N Z E NZE No LO D D Yes L o w Central NZE NZE Yes PLENTY Yes Low Low N Z E D e l a y No

1. (Planned coal/expected need) relatively small. 2. " " " " " c entral. 3. " " " " " large. 4. All scenarios are solved for a discount rate of 10%, some are solved for 4% as well. 145

3. THE SIMULATION MODEL: FEATURES

(1) 'Time'-staged. The horizon covers one peak and one off- peak demand segment each quarter from 1/4/1981 to 31/3/1996. i.e. 120 'periods' in all. (2) Two regions. The North and South Islands are separately represented, with the DC-link allowing energy transfers. (3) Hydro-system. Generation in each period is determined from 'rules' as a function of start of period storage. Random inflows in the period determine end of period storage and may influence feasibility of generation. (4) Thermal Stations. Demand not met by hydro generation is supplied from thermal stations, loaded in merit order, limited by capacity. (5) HPS. As well as in merit order, generation may occur to keep the stockpile below the capacity specified by the coal policy. Coal shortage may limit generation. (6) Underground mining. Production in each period is specified by the coal policy. (7) Opencast mining. Stripping and recovery are endogenously determined. If prestripped opencast stocks fall below a level specified in the coal policy, stripping 'contracts' are let. Opencast coal is used only when other supplies are inadequate. (8) Costs. The cost calculated for minimisation is the total discounted cost of all thermal fuels used (except HPS coal), plus coal stockpile establishment and operating costs, plus mining costs including capital charges associated with new sources.

4. THE SIMULATION MODEL: USE

Each iteration through the 120 periods produces a possible set of physical outcomes for the coal/power system, and a con­ sequent cost, which depends on the coal policy, scenario, and hydro inflows. The average cost over many iterations with different randomly generated inflows is therefore the expected cost of a policy, conditional on a scenario. For each scenario in turn the policy with minimum expected cost is found by suc­ cessive policy improvement using an interactive computer based routine. The user inputs a policy change, simulates to derive costs and so-on in sequence, until no further improvement can be found. Some physical quantities are very good indicators of directions for policy improvement, the most important being generation lost at HPS through coal shortage, and coal burnt at HPS to reduce stockpile size. Improvement is rapid, due to the 'well-behaved' nature of the underlying cost function. 146

5. HYDRO RULES

The use of hydro system storage reservoirs to 'shift' inflows through time can have a large impact on electricity generating costs. Optimal scheduling of reservoir releases is a widely con­ sidered problem of operations research. It was treated here by developing a set of hydro release rules for each scenario by stochastic dynamic programming.

The formulation necessarily largely follows the simulation model. Stages are periods, and the two state variables are the North and South Island start of period hydro reservoir storage levels (suitably discretized). The two decision variables are the North and South Island target hydro generation levels. The objective is to minimise total discounted fuel costs. Thermal stations fill the demand not met by hydro generation, and are loaded in merit order. Here, HPS is treated as any other thermal station, all coal needed is assumed to be available, and coal used incurs marginal costs as with other fuels.

Solution by stochastic dynamic programming produces, in each scenario/period/state, three numbers; the 'optimal' North Island target hydro generation, the 'optimal' South Island target hydro generation, and the expected total discounted cost (of generating electricity) from that state and period to the horizon end. A sample set of results appears as Table 2.

6. BEST POLICIES

As is seen in Table 3, the best policies found are fairly similar for all but low demand cases. These cases aside, HPS is extremely close to merit order operations under the best policies. This indicates that coal supply should be planned to meet generation needs, and not vice-versa. In the low demand cases HPS is supply driven. Here the slowest Huntly West buildup rate considered initially produces more coal than merit order operations require, and the surplus cannot economically be stored until required.

Best policies are insensitive to the assumptions and para­ meters of the model, as is seen for holding cost in Figure 1. The cost of oversupply of coal is far less than the cost of under­ supply as can be seen in Table 4. This may be sufficient inform­ ation for long range general planning recommendations; i.e. plan for a very large stockpile, set the Huntly West buildup rate as planned or a little slower, and supply additional needs from open­ cast mines with a large prestripped stock. 14 7

T a b l e 2: sdp results for A-M-J quarter/offpeak/1981/central. where: NGEN = "optimal" N.I. target generation given storage state (gWh) SGEN = "optimal" S.I. target generation given storage state (gWh) ECOST= expected cost, given storage state, to horizon end ($M)

NORTH ISLAND S T A R T OF S.I. SOP 'OPTIMAL' PERIOD STORAGE (gWh) STORAGE VALUE 0 153 307 460 614 767

0 N G E N 1000 1100 1100 1200 1400 1500 SGEN 1900 1900 1900 1800 1600 1700 ECOST 265.679 260.099 2 5 5 . 7 7 9 2 5 3 . 1 6 6 251. 1 9 0 249.506

233 N G E N 900 900 1100 1200 1400 1500 SGEN 2100 2100 1900 1900 1800 1900 ECOST 2 57.716 254.322 252 . 0 3 8 2 5 0 . 2 8 7 248.615 246.951

466 NGEN 800 900 1100 1200 1400 1500 SGEN 2200 2100 2100 2100 2000 2000 ECOST 253.062 251.108 2 4 9 . 4 4 1 2 4 7 . 7 4 5 246.112 244.525

800 NGEN 800 900 1100 1200 1400 1500 SGEN 2300 2300 2300 2300 2100 2000 ECOST 250.278 248.541 246. 9 0 9 2 4 5 . 2 5 2 2 43.807 242 . 6 1 3

933 NGEN 800 900 1100 1200 1400 1500 SGEN 2500 2600 2400 2300 2200 2100 ECOST 2 47.727 246.031 244. 4 7 4 2 4 3 . 1 4 5 242. 1 0 3 241.089

1166 NGEN 800 900 1000 1200 1300 1500 SGEN 2700 2600 2500 2400 2300 2100 ECOST 245.243 2 43.729 2 4 2 . 6 0 6 24 1 . 5 3 6 240.671 239. 9 9 0

1399 NGEN 700 900 1000 1200 1300 1500 SGEN 2800 2700 2600 2400 2300 2100 ECOST 243.146 242.024 241.077 2 4 0 . 2 7 9 2 39.726 239.338

16 i2 NGEN 800 900 1000 1200 1300 1400 SGEN 2800 2700 2600 2400 2300 2200 ECOST 241.579 240.620 239. 9 6 7 23 9 . 4 9 7 239.205 239.036

186 5 NGEN 800 900 1000 1100 1200 1400 SGEN 2800 2700 2600 2 500 2400 2200 ECOST 240.362 239.706 2 39.323 2 3 9 . 1 0 6 2 38.979 238.925

2099 NGEN 800 800 900 1000 1200 1300 SGEN 2800 2800 2700 2600 2400 2300 ECOST 239.607 239.202 239.032 2 38.947 2 38.905 238.892

2332 NGEN 800 800 900 1000 1200 1300 SGEN 2800 2800 2700 2600 2400 2300 ECOST 239.240 238.982 238 . 9 2 0 2 3 8 . 8 9 6 2 38.887 238.884

2565 NGEN 800 800 900 1000 1200 1300 SGEN 2800 2800 2700 2600 2400 2 300 ECG.yr 239.099 238.910 238. 3 9 0 2 3 8 . 8 8 6 238.884 238.883 148

Table 3: Best Policies by Scenario

SCENARIO NAME M I N I M U M STOCKPILE* FIRST YEAR TARGET* UNDERGROUND IN T A B L E 1 FUEL & COAL CAPACITY OPENCAST OPENCAST PRODUCTION Sc C A P I T A L AVAILABLE C.f. PLA N S C O S T $(M) PAUCITY 515 260 0 1987 1500 As planned HIDD 464 2200 1987 1500 A little slo w e r HDELY 377 2200 1907 1400 A little slo w e r HIOIL 369 2 6 0 0 1988 1400 A little s l o w e r CENTRAL 358 2600 1988 1400 A little slo w e r TDELY 355 2200 1988 1400 A litt l e slower LOOIL 349 2200 1988 1400 S l o w e r NSMID 311 2200 1989 1400 S l o w e r LODD 265 800 1987 800 Much slower PLENTY 247 200 1989 600 Much slower

* t h o u s a n d s o f t o n n e s Expected total discounted c o s t . $ (M) 395'

600 1000 1400 1800 2200 2600 3000 3400 Stockpile Capacity. Thousands of tonnes 149

Table 4: Cost of Policies in various situations

S I T U A T I O N 1 1 BE S T ’ POLICY USED2 PLENTY LODD CENTRAL HIDD PAUCITY

PLENTY 2 4 7 3 269 433 715 1010 LODD 252 265 443 731 1006 CENTRAL 287 292 358 501 640 HIDD 284 291 359 464 525 PAUCITY 299 305 367 468 515

1. see T a b l e 1. 2. see T a b l e 3. 3. $ (M ) .

7. AN OVERALL STRATEGIC BEST

A strategy fixes current decisions, with future decisions dependant on intervening events and outcomes. Current decisions are the underground production,stockpile capacity, and first year of opencast stripping. Prestripped opencast target levels can be decided in future. Possible future events are given by scenarios. The expected cost of each strategy/scenario is found as the minimum expected cost over opencast target levels.

The best strategy is sought. Low coal supply options are not considered because of the high cost of undersupply as shown in Table 4. Strategies derived from the remaining best policies, and their costs under the scenarios, appear in Table 5.

Some strategies may be eliminated because another strategy dominates them, i.e. has a lesser cost in every outcome. In Table 5 we see that

2 and 4 are dominated by 7 5, 6, and 8 are 'nearly' dominated by 7 1 is 'nearly' dominated by 3.

The remaining strategies, 3 and 7, differ on underground product­ ion. Strategy 3 is minimax best, but minimum expected cost is a more appropriate decision criterion because high regret options have already been discarded. Specification of prior probabilities is avoided as follows:

Define: P(A) = prior probability on occurrence of scenario A. Assume: P (!7B) + P(HD) + P(AC) + P(LD) + P(LB) - 1.0

vhr-re scenario names are abbreviated as in Table 5. Then the expected cost of strategy 3 is less .than the expected cost of Table 5: Costs of Possible Strategies 150

T h o u s a n d s 151

strategy 7 if:

P(HD)(469-464)+P(AC)(368-359)+P(LD)(298-287) + P(LB) (289-278)

1 1 ( 1 - P ( U B ) ) < 10 (P(UB) )

Therefore 3 is better than 7 if P(UB) > 11/21.

By similar reasoning:

7 is better than 3 if P(UB) < 1/3 7 or 3 may be the better if 1/3 < P(UB) < 11/21.

It seems very likely that the prior probability of occurrence of the upper bound coal need scenario is less than 1/3,so strategy 7 minimises expected cost and is the overall strategic best option. It follows that current steps necessary to achievement of the coal operations of strategy 7 should be taken, with actual achievement depending on intervening outcomes and further rounds of planning c y c l e s .

REFERENCE

Baker, W. R. Optimal Coal Mining and Stockpiling Strategies for the Huntly Power Station, unpublished Master of Science Thesis, University of Canterbury, 1^82. 152