SPS Optimisation PhD Steering Group

12th January 2018 Introduction

• Thames Water Consumes Renewables play an important role in the 1,200GWh P/A • Their Renewables contribute to an water industries move towards a reduced offset of 156GWh Carbon Footprint. • TOTEX Model focusing more on Whole-life costs However, should the concept of generating ‘more’ to meet demands be revised?

Should there be a shift in directive?

How can we control our existing asset base more efficiently?

This PhD Research is focusing on optimising the automation and control surrounding sewage pumping stations with a view to reducing the energy required to drive the assets.

2 Recap from Last SGM

December 2016

• Discussion on Literature Review • High Level Review on ANN’s and Genetic Algorithms • UKWIR Report on 5x ANN Projects to Predict CSO Spills – reduced requirement on complex models • Research at Sheffield showed good results with Genetic Algorithms but was limited to one site and one week of data • Changes to the Live System Highly Unlikely without Proof of Concept • ICMLive Infoworks Model to be Utilised

3 Recap – from K-Means Algorithm Bordon Catchment Identified • Use of K-Means Algorithm Identified Bordon Catchment • This Catchment had most ‘Optimised Multismart’s’ at time of analysis • Potential for maximum control change with minimum investment in hardware.

4 Initial Model Review

Consultation Session and 1-day Application Training with Pascal Lang from the Modelling Team enabled the research to continue with trialling Bordon

• Opportunity to trial multiple profiles/start stop conditions on the model • Ability to forecast spills in a safe environment

• Start / Stop Level Issues • Presented in meters above datum. • Not validated against the real system, lots of assumption.

Decision made to Survey the Catchment

5 Site Surveys

33 Sites Visited Varying levels of FOG deposits in the catchment

Tunbridge Lane SPS

6 Site Surveys Ordnance Survey Team • Select Surveys brought in to validate 33 SPS cover levels • All measurements then converted to OD and updated in the model. • The model was now reflective of the actual catchment • There were some major discrepancies between the previous settings and the new.

7 Head/Discharge Curves

• Pump Tag Data Captured during Site The Hydraulic Model has the ability to Survey program the individual pumps in the • Manufacturers Literature used to extract catchment with head/discharge head discharge curves for each pump type. parameters. • Each pump updated in model with accurate head discharge curves according to the current Bordon catchment • Model now ready for test runs.

8 Rainfall

July 2017 • Particularly wet month chosen to create design rainfall object • Inputted into model @ 1min Sample rate over 1month • Huge Data Blocks now emerging sparking potential strategy to move towards cloud based modelling?

HEATHROW HURN AVERAGE yyyy mm tmax tmin af rain sun yyyy mm tmax tmin af rain sun yyyy mm tmax tmin af rain degC degC days mm hours degC degC days mm hours degC degC days mm 2016 1 9.5 3 5 74.8 54.8# 2016 1 10 2.1 10 190.8 66.7# 2016 1 9.75 2.55 7.5 132.8 2016 2 9.4 2.9 5 43.8 86.8# 2016 2 9.6 2.7 10 63.2 64.6# 2016 2 9.5 2.8 7.5 53.5 2016 3 10.7 3.2 3 72.8 123.3# 2016 3 10.9 1.4 11 105.6 132.4# 2016 3 10.8 2.3 7 89.2 2016 4 13.5 4.9 0 47.2 170.0# 2016 4 12.9 3.2 6 56.8 166.2# 2016 4 13.2 4.05 3 52 2016 5 19 9.7 0 60.4 192.7# 2016 5 18 7.5 2 58.8 202.9# 2016 5 18.5 8.6 1 59.6 2016 6 20.7 12.7 0 93.4 101.7# 2016 6 19.6 11.4 0 85.8 116.0# 2016 6 20.15 12.05 0 89.6 2016 7 24 14.5 0 16 182.8# 2016 7 22.4 12.6 0 12 190.0# 2016 7 23.2 13.55 0 14 2016 8 24.7 14.6 0 21.6 201.4# 2016 8 23.1 12.3 0 41 230.2# 2016 8 23.9 13.45 0 31.3 2016 9 22.4 13.7 0 42.2 122.1# 2016 9 20.7 12 0 85.2 125.8# 2016 9 21.55 12.85 0 63.7 2016 10 15.9 8.7 0 21.6 105.6# 2016 10 15.5 6.1 0 56.6 125.7# 2016 10 15.7 7.4 0 39.1 2016 11 10.5 3.8 3 86.4 77.4# 2016 11 11.1 2 9 104.4 97.1# 2016 11 10.8 2.9 6 95.4 2016 12 10.2 3.4 7 10.4 55.1# 2016 12 10.3 2.1 10 32.2 55.8# 2016 12 10.25 2.75 8.5 21.3 2017 1 7.6 0.7 15 60.2 64.5# Provisional 2017 1 8.1 -0.1 17 103.4 74.1# Provisional 2017 1 7.85 0.3 16 81.8 2017 2 10 4.4 1 38.2 47.8# Provisional 2017 2 9.9 3.7 7 47.4 56.0# Provisional 2017 2 9.95 4.05 4 42.8 2017 3 14.1 6.6 0 25.8 116.3# Provisional 2017 3 13.2 5.5 1 32.4 107.0# Provisional 2017 3 13.65 6.05 0.5 29.1 2017 4 15.8 5.9 0 4.6 186.2# Provisional 2017 4 15 3.2 5 4.8 193.8# Provisional 2017 4 15.4 4.55 2.5 4.7 2017 5 19.8 10.4 0 64.8 164.8# Provisional 2017 5 18.3 8.3 0 69.4 185.9# Provisional 2017 5 19.05 9.35 0 67.1 2017 6 24 13.9 0 46.4 204.3# Provisional 2017 6 21.9 11.4 0 52.6 207.3# Provisional 2017 6 22.95 12.65 0 49.5 2017 7 23.8 14.9 0 90 178.0# Provisional 2017 7 22.2 12.8 0 76.8 185.4# Provisional 2017 7 23 13.85 0 83.4 2017 8 22 13.5 0 58.6 162.9# Provisional 2017 8 21.4 12.7 0 66.1 217.6 Provisional 2017 8 21.7 13.1 0 62.35 2017 9 19.2 11 0 59 120.1# Provisional 2017 9 18.1 11.2 0 90.3 * 129.4* Provisional 2017 9 18.65 11.1 0 74.65 2017 10 17.1 10.3 0 11.4 91.2* Provisional 2017 10 16.5 10 0 43.8 * 92.6* Provisional 2017 10 16.8 10.15 0 27.6 2017 11 11.1 4.5 2 34.2 66.7# Provisional 2017 11 11.9 4.6 2 74.8 * 104.6* Provisional 2017 11 11.5 4.55 2 54.5 9 Energy - Model Product Limitations • Some issues with product support • Eventually discovered that the software does not calculate power • New approach required catchment power to be calculated via Downstream Flow and Head • As Pumps only run for short durations, the model had to be set at high frequency sampling rates. • Model files generated are up to 100GB per run • For 1 month, this equated to 483841 lines of data per pump in CSV format Power Calculations • Use model to calculate mechanical energy in hydrostatic load (fluid on open circuit)

푃푓푙 = 푄 . 휌. 퐻. (9.81) • Degraded energy expressed by the output of the pump (power at the shaft of the pump)

• Q = Flow in m3/s. 푃푓푙 푃 = • p = density of the liquid in kg/m3. 푚푒푐 • H = Hydraulic load in meter of water. 푅푣. 푅푡 • 9.81 = Average Intensity of gravity. • Pmec = Mechanical power necessary to the pump. • Pfl = Power transmitted to the fluid. • Rv = Output of the ventilator. • Rt = Output of the transmission. 10 Model Validation

Bordon July 17 KWHr Outliars Removed

N-Power Site Name Meter Model Variance Average Variance Across Model % Apollo Drive (Bordon) SPS 1,690 341.9778 -79.7646 37.01951478 SPS 1,016 14945.5 1371.448 Ash Grove () SPS 54 5.013251 -90.7162 Court () SPS 402 12.46703 -96.8987 Alan Jeffery from Business Chalet Hill (Bordon) SPS 365 1306.149 257.8491 Chapel Gardens (Lindford) SPS 221 98.35477 -55.4956 Systems Management Churt SPS 5,254 5228.312 -0.48324 provided energy data taken SPS 3,142 6003.778 91.0814 Cypress Road (Bordon) SPS 574 168.1145 -70.7118 from the N-Power meters. Griggs Green (Liphook) SPS 1,620 495.2463 -69.4292 Hamilton Close (Bordon) SPS 48 6.368246 -86.7328 Headley Mill (Bordon) SPS 707 2344.538 231.6178 This is what the business is Heatherlands () SPS 272 Hogmoor Road (Whitehill) SPS 1,169 1229.496 5.175056 billed against. It is the most Kingsley Common (Bordon) SPS 815 399.7461 -50.9514 accurate for consumed kWHr Lions Field (Oakhanger) SPS 88 46.92733 -46.6735 Malthouse Meadows (Liphook) SPS 229 45.26516 -80.2336 and was used to validate the Maple Leaf Drive (Bordon) SPS 48 Marsh Close (Bordon) SPS 118 48.80282 -58.6417 model runs Monument Chase (Whitehill) SPS 129 34.4899 -73.2636 Oakhanger SPS 388 610.4727 57.33832 Passfield (Bramshott) SPS 3,688 Some sites showing really Royal Drive (Bordon) SPS 273 Stoney Bottom () SPS 1,654 close comparisons but The Meadows (Churt) SPS 42 33.1553 -21.0588 others are way out. Tulls Lane () SPS 394 24.84913 -93.6931 Tunbridge Lane (Bramshott) SPS 72 34.54696 -52.0181 Walldown Road (Whitehill) SPS 1,087 571.4166 -47.4318 Average variance across Warren Close (Whitehill) SPS 393 314.7663 -19.9068 Western (Bordon) SPS 1,209 2916.048 141.195 model and power data is 37% Whitmore Vale (Grayshott) SPS 1,038 456.2021 -56.0499 Woolmer SPS 1,236 826.8384 -33.1037 28,198 38548.84 36.70577 11 Model – Scenario 1

Bordon July 17 KWHr Difference from Site Name Model Base 50mm Increase on all Pump Starts Apollo Drive (Bordon) SPS 336.7404 5.2374 Arford SPS 14853.43 92.07 Ash Grove (Liphook) SPS 4.777083 0.236168 Bramshott Court (Passfield) SPS 12.22097 0.24606 Chalet Hill (Bordon) SPS 1294.442 11.707 • A standard 50mm increase was applied Chapel Gardens (Lindford) SPS 97.32334 1.03143 to all start levels within the model. Churt SPS 5215.621 12.691 Conford SPS 6023.167 -19.389 • The model was rerun and the power Cypress Road (Bordon) SPS 166.2786 1.8359 recalculated from the downstream head Griggs Green (Liphook) SPS 459.8011 35.4452 Hamilton Close (Bordon) SPS 5.643541 0.724705 and flow. Headley Mill (Bordon) SPS 2338.589 5.949 Heatherlands (Headley Down) SPS 0 • No spills reported from model, and this Hogmoor Road (Whitehill) SPS 1151.835 77.661 was based on the same wet July where Kingsley Common (Bordon) SPS 397.2326 2.5135 Lions Field (Oakhanger) SPS 46.72293 0.2044 83.4mm rainfall landed Malthouse Meadows (Liphook) SPS 44.25929 1.00587 • 325kWhr over 1month saving = £38.35. Maple Leaf Drive (Bordon) SPS 0 Marsh Close (Bordon) SPS 47.64949 1.15333 Monument Chase (Whitehill) SPS 33.49926 0.99064 • Not the most exciting result. Oakhanger SPS 600.775 9.6977 Passfield (Bramshott) SPS 0 Royal Drive (Bordon) SPS 0 Stoney Bottom (Grayshott) SPS 0 The Meadows (Churt) SPS 33.10575 0.04955 Tulls Lane (Standford) SPS 24.11175 0.73738 Tunbridge Lane (Bramshott) SPS 33.79961 0.74735 Walldown Road (Whitehill) SPS 568.5372 2.8794 Warren Close (Whitehill) SPS 304.7981 9.9682 Western (Bordon) SPS 2884.559 31.489 Whitmore Vale (Grayshott) SPS 454.5753 1.6268 Woolmer SPS 789.8684 36.97 12 38223.37 325.47 Going Forward

• Is the model accurate enough, do we need What are the next steps for this research? further validation? Where does the industrial sponsor wish to • Can we use it in its current state to justify see this go? a production field trial? • Option to try Deterministic / Genetic optimisation techniques? • Is Bordon the right catchment or should focus on larger areas? It only costs £47k per annum for the 33 sites, even a 20% reduction would not cover the costs of deployment. • Where is the business in terms of its SPS optimisation strategy? Is there budget?

13 Questions???

14