A Detailed Comparison of Meta- Heuristic Methods for Optimising Wave Energy Converter Placements Mehdi Neshat, Bradley Alexander, Markus Wagner, Yuanzhong Xia GECCO ‘18 Optimisation and Logistics Group Slide 1
Growth of Renewable Energy • Renewable energy – (wind and solar) are now the cheapest form of new-build power generation. – Solar contracts ~US 2c/kWh • (Saudi Arabia – 1.79c kWh (the national Abu Dhabi – Jan 2018)). • Growing level of investment – Global investment US $263 billion in 2016 (source IRENA, Jan 2018) • GECCO ‘18 Optimisation and Logistics Group Slide 2
Problem Source: NEM-Watch • Plenty of renewables… – South Australia • 15th July, 2018 GECCO ‘18 Optimisation and Logistics Group Slide 3
Problem • Plenty of renewables… – South Australia • 15th July, 2018 Demand ~1.8GW GECCO ‘18 Optimisation and Logistics Group Slide 4
Problem • Plenty of renewables… – South Australia • Morning 15th July, 2018 Renewable’s Share >80% GECCO ‘18 Optimisation and Logistics Group Slide 5
Problem • ..but intermittent. Source: Open-NEM GECCO ‘18 Optimisation and Logistics Group Slide 6
Problem • ..but intermittent. 15 th July GECCO ‘18 Optimisation and Logistics Group Slide 7
Problem • ..but intermittent. But what about here! GECCO ‘18 Optimisation and Logistics Group Slide 8
Problem • current role of storage Note the battery (world’s largest) GECCO ‘18 Optimisation and Logistics Group Slide 9
Possible Solutions • Need to smooth and/or time-shift generation • Alternatives – More-connected Grid? • Helps but expensive – Pumped Hydropower? • Need water and hills – Batteries? • Fast and efficient but still too small.. – Gas Peaking? • Can be expensive + carbon emissions. GECCO ‘18 Optimisation and Logistics Group Slide 10
Wave Energy Several potential advantages • Low correlations with local wind • – Correlated with distant winds High Capacity Factor • – Over 70% High Energy Density • – Over 60 times that of solar per m 2 Neshat, et. al. GECCO ‘18 Slide 11 Optimisation and Logistics Group
Wave Energy Converters GECCO ‘18 Optimisation and Logistics Group Slide 13
Wave Energy Converters buoys GECCO ‘18 Optimisation and Logistics Group Slide 14
Wave Energy Converters buoys GECCO ‘18 Optimisation and Logistics Group Slide 15
Wave Energy Converters buoys Problem: Place buoys to maximise p GECCO ‘18 Optimisation and Logistics Group Slide 16
Wave Energy Converters buoys Problem: Place buoys to maximise p GECCO ‘18 Optimisation and Logistics Group Slide 17
Buoy Placement is Non-trivial • Question: Why not just place buoys all in a line, as far apart as possible? • Constructive Interference – You can get more energy offtake by placing buoys close to each other. – But not too close! – Depends on local wave conditions. • And, interactions become complex as more buoys are added. • And, the size of wave farms is limited. GECCO ‘18 Optimisation and Logistics Group Slide 18
Other Work Authors Methods Results Gaps ref (year) A. D. De Evaluating different fixed shape A triangular shape with different Limited [1] Andrés et. models by various wave directions. wave directions and a square shape shape of al (2014) with a unidirectional wave are the array best. C.J.Sharp Tuned GA Optimal 5-buoy layout with q- Simpler [2] (2015) factor=1.024 (Best) model, >37000 (evaluations) Wu et. al (1+1)EA and (2+2)CMA-EA Optimal 25-buoy layout (q-factor= One wave [3] (2016) 0.9 and 100-buoy layout (q-factor= frequency 0.74. and one wave direction [11] Sharp et. al. Tuned GA Optimises cost and energy output – 5-buoy (2018) discrete grid placement – high and layout, low intervals. GECCO ‘18 Optimisation and Logistics Group Slide 19
The Fitness Function • The fitness function is computationally intensive. • Each evaluation calculates: For each buoy For each wave frequency For each wave direction (lookup wave-height distribution) For each other buoy in the farm Model Hydrodynamics and estimate energy • Can take up to 9 minutes for a full evaluation – 16 buoys on 12 cores GECCO ‘18 Optimisation and Logistics Group Slide 20
Constraints • 1) Upper bound for the farm area • 2) safe distance between buoys GECCO ‘18 Optimisation and Logistics Group Slide 21
Optimisation Targets 4 Buoy Farm 16 Buoy Farm GECCO ‘18 Optimisation and Logistics Group Slide 22
Optimisation Setup • 13 CPU machine • Depending on optimisation meta-heuristic, either: – Evaluate individual layouts in parallel – Evaluate wave frequencies in parallel GECCO ‘18 Optimisation and Logistics Group Slide 23
Meta-Heuristics (1) GECCO ‘18 Optimisation and Logistics Group Slide 24
Meta-Heuristics (1) Random Search – place all n buoys simultaneously GECCO ‘18 Optimisation and Logistics Group Slide 25
Meta-Heuristics (1) Partial Evaluation – estimate fitness on random subset of frequencies – saves time! Duc-Cuong Dang and Per Kristian Lehre. 2014. Evolution Non elitist! under partial information. In Proceedings of the 2014 Annual Conference on Genetic and Evolutionary Computation (GECCO '14). ACM, New York GECCO ‘18 Optimisation and Logistics Group Slide 26
Meta-Heuristics (1) TDA – used for wind-turbine placement. Wagner, Markus & Day, Jareth & Neumann, Frank. (2012). A Fast and Effective Local Search Algorithm for Optimizing the Placement of Wind Turbines. Renewable Energy. 51 ( 2013), 64–70 GECCO ‘18 Optimisation and Logistics Group Slide 28
Meta-Heuristics (1) CMA ES – custom and from previous work in field. Junhua Wu, Slava Shekh, Nataliia Y Sergiienko, Benjamin S Cazzolato, Boyin Ding, Frank Neumann, and MarkusWagner. Fast and effective optimisation of arrays of submerged wave energy converters. In GECCO 2016, ACM, 1045–1052. GECCO ‘18 Optimisation and Logistics Group Slide 29
Meta-Heuristics (1) Differential Evolution GECCO ‘18 Optimisation and Logistics Group Slide 30
Meta-Heuristics (1) 1+1 EA’s with different mutation strategies GECCO ‘18 Optimisation and Logistics Group Slide 31
Meta-Heuristics (2) GECCO ‘18 Optimisation and Logistics Group Slide 32
Meta-Heuristics (2) Buoy at-a-time placement – starts fast, finishes slow. GECCO ‘18 Optimisation and Logistics Group Slide 33
Meta-Heuristics (2) Random local neighbourhood search GECCO ‘18 Optimisation and Logistics Group Slide 34
Meta-Heuristics (2) Random placement + downhill search on all buoys. GECCO ‘18 Optimisation and Logistics Group Slide 35
Meta-Heuristics (2) Alternate placement and downhill search GECCO ‘18 Optimisation and Logistics Group Slide 36
Meta-Heuristics (2) Smart offsets for placement of next buoy – different local search and refinements. GECCO ‘18 Optimisation and Logistics Group Slide 37
Why Smart? • Local landscape (impact of adding second buoy)… GECCO ‘18 Optimisation and Logistics Group Slide 38
Comparing Algorithm Performance • Methods have different number of wave frequency evaluations and different placement strategies. • Not fair to measure just in terms of evaluations. • Most practical measure is the performance of the best layouts for each algorithm – dedicated machine – with a fixed runtime • Runtime – 3 days – 13 processors GECCO ‘18 Optimisation and Logistics Group Slide 39
Results: Energy – 4 Buoy layout Many high-performing heuristics – capacity factor > 1 GECCO ‘18 Optimisation and Logistics Group Slide 41
Layout– 4 Buoys Different heuristics – similar shaped layout GECCO ‘18 Optimisation and Logistics Group Slide 43
Results: Energy – 16 Buoy layout More challenging and constrained problem GECCO ‘18 Optimisation and Logistics Group 44
Results: Energy – 16 Buoy layout Smart local search wins GECCO ‘18 Optimisation and Logistics Group 45
…But Partial Evaluation is OK GECCO ‘18 Optimisation and Logistics Group 46
…but Partial Evaluation is OK Small populations or frequencies do well GECCO ‘18 Optimisation and Logistics Group 47
Partial Evaluation Traces – 16 buoys GECCO ‘18 Optimisation and Logistics Group Slide 48
Partial Evaluation Traces – 16 buoys Small numbers of frequencies converge better. GECCO ‘18 Optimisation and Logistics Group Slide 49
Traces for Population Based Methods GECCO ‘18 Optimisation and Logistics Group Slide 50
Traces for 1+1EAs and Smart Placements GECCO ‘18 Optimisation and Logistics Group Slide 51
Best 16-buoy layout The best layout of LS_1+NM_2Dfor 16 buoy research. The area size is 566m x 566m, the q- factor=0.956, total power output 7608600W, GECCO ‘18 Optimisation and Logistics Group Slide 52
Conclusions • Wave Energy Farm Design is a challenging problem. • The computational budget is very limited • Any informed tricks that can be used to help reduce this budget can help. GECCO ‘18 Optimisation and Logistics Group Slide 53
Recommend
More recommend
