AUTHOR=Gao Yanna , Dong Hong , Hu Liujun , Zeng Fanhong , Gao Yuqun , Huang Zhuonan , Wang Shaohua 

TITLE=Deep reinforcement learning for multi-objective location optimization of onshore wind power stations: a case study of Guangdong Province, China

JOURNAL=Frontiers in Energy Research

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/energy-research/articles/10.3389/fenrg.2025.1596471

DOI=10.3389/fenrg.2025.1596471

ISSN=2296-598X

ABSTRACT=IntroductionWind energy development faces challenges such as low utilization of wind resources, underdevelopment of suitable areas, and imbalanced electricity demand coverage. To address these issues, this study formulates a multi-objective maximal covering location problem (MO-MCLP) for onshore wind power station (OWPS) siting, aiming to improve resource utilization, expand development in promising regions, and balance demand coverage in spatial planning.MethodsA MO-MCLP model is developed that simultaneously maximizes wind energy utilization, promotes development in suitable areas, and balances electricity demand coverage. To solve this model at large scale, a deep reinforcement learning (DRL) algorithm is designed and implemented. The DRL approach is benchmarked against a traditional optimization implementation using the Gurobi solver. Computational experiments focus on wind-rich coastal regions of Guangdong Province, evaluating both solution quality (coverage and utilization metrics) and computational efficiency under varying problem sizes.ResultsThe DRL algorithm achieves objective values comparable to or better than those from the Gurobi-based method, while substantially reducing computation time for large problem instances. As the number of candidate sites and demand points increases, DRL demonstrates superior scalability. In the Guangdong case study, DRL attains similar or improved coverage and utilization within a fraction of the runtime required by Gurobi, enabling faster iteration for scenario analysis.DiscussionThe findings indicate that DRL offers an efficient alternative to traditional solvers for complex spatial optimization in wind farm siting. Faster computation and better scalability facilitate exploration of multiple planning scenarios, sensitivity analyses, and rapid decision support under practical time constraints. Integrating richer environmental and socioeconomic data, extending to multi-stage planning, or combining DRL with heuristic solvers may further enhance performance. Overall, the MO-MCLP model with DRL solution provides actionable insights for sustainable energy infrastructure planning by delivering high-quality site allocations efficiently.