Wind farm layout optimization using self-informed genetic algorithm with information guided exploitation

Wind energy which is known for its cleanliness and cost-effectiveness has been one of the main alternatives for fossil fuels. An integral part is to maximize the wind energy output by optimizing the layout of wind turbines. In this paper, we first discuss the drawbacks of Conventional Genetic Algorithm (CGA) by investigating into the implications of crossover and mutation steps of CGA for the wind farm layout problem, which explains why CGA has a higher possibility of convergence to a suboptimal solution. To address the limitations of CGA, we propose novel algorithms by incorporating the self-adaptivity capability of individuals, which is an essential step observed in the natural world, called Adaptive Genetic Algorithm (AGA) and Self-Informed Genetic Algorithm (SIGA). To be specific, the individual's chromosomes in a population will conduct a self-examination on the efficiency of all the wind turbines, and thus gaining self-awareness on which part of the solution is currently the bottleneck for further improvement. In order to relocate the worst turbine, we first propose to relocate the worst turbine randomly with AGA, and then an improved version called SIGA is developed with information guided relocation to find a good location using a surrogate model from Multivariate Adaptive Regression Splines (MARS) regression based on Monte Carlo Simulation. Extensive numerical results under multiple wind distributions and different wind farm sizes illustrate the improved efficiency of SIGA and AGA over CGA. In the end, an open-source Python package is made available on github (https://github.com/JuXinglong/WFLOP_Python).