TY - JOUR
T1 - Wind farm layout optimization using adaptive evolutionary algorithm with Monte Carlo Tree Search reinforcement learning
AU - Bai, Fangyun
AU - Ju, Xinglong
AU - Wang, Shouyi
AU - Zhou, Wenyong
AU - Liu, Feng
N1 - Publisher Copyright:
© 2021 Elsevier Ltd
PY - 2022/1/15
Y1 - 2022/1/15
N2 - Recent years have witnessed an enormous growth of wind farm capacity worldwide. Due to the wake effect, the velocity of incoming wind is reduced for the wind turbines in the downwind directions, thus causing discounted power generation in a wind farm. Previously, a self-informed adaptivity mechanism in evolutionary algorithms was introduced by the authors, which is inspired by the individuals’ self-adaptive capability to fit the environment in the natural world, where relocating the worst wind turbine with a surrogate model informed mechanism was found to be effective in improving the power conversion efficiency. In this paper, the exploitation capability in the adaptive genetic algorithm is further improved by casting the relocation of multiple wind turbines into a single-player reinforcement learning problem, which is further addressed by Monte-Carlo Tree Search embedded within the evolutionary algorithm. In contrast to the moderate improvements of the authors’ previous algorithms, significant improvement is achieved due to the enhanced algorithmic exploitation. The new algorithm is also applied to solve the optimal layout problem for a recently approved wind farm in New Jersey, and showed better performance against the benchmark algorithms.
AB - Recent years have witnessed an enormous growth of wind farm capacity worldwide. Due to the wake effect, the velocity of incoming wind is reduced for the wind turbines in the downwind directions, thus causing discounted power generation in a wind farm. Previously, a self-informed adaptivity mechanism in evolutionary algorithms was introduced by the authors, which is inspired by the individuals’ self-adaptive capability to fit the environment in the natural world, where relocating the worst wind turbine with a surrogate model informed mechanism was found to be effective in improving the power conversion efficiency. In this paper, the exploitation capability in the adaptive genetic algorithm is further improved by casting the relocation of multiple wind turbines into a single-player reinforcement learning problem, which is further addressed by Monte-Carlo Tree Search embedded within the evolutionary algorithm. In contrast to the moderate improvements of the authors’ previous algorithms, significant improvement is achieved due to the enhanced algorithmic exploitation. The new algorithm is also applied to solve the optimal layout problem for a recently approved wind farm in New Jersey, and showed better performance against the benchmark algorithms.
KW - Adaptive genetic algorithm
KW - Evolutionary computations
KW - Monte-Carlo Tree Search
KW - Reinforcement learning
KW - Wind farm layout optimization
UR - http://www.scopus.com/inward/record.url?scp=85120496487&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85120496487&partnerID=8YFLogxK
U2 - 10.1016/j.enconman.2021.115047
DO - 10.1016/j.enconman.2021.115047
M3 - Article
AN - SCOPUS:85120496487
SN - 0196-8904
VL - 252
JO - Energy Conversion and Management
JF - Energy Conversion and Management
M1 - 115047
ER -