Research on Substation Siting Optimization Based on Deep Reinforcement Learning

Substation siting is a critical component of power system planning, involving multi-objective optimization problems such as service coverage, economic cost, and geographical adaptability. To address the limitations of traditional methods in adapting to complex environments and dynamic adjustments, this paper proposes an intelligent siting optimization method based on Deep Reinforcement Learning (DRL). The method employs an Actor-Critic architecture to construct an environmental model in a two-dimensional gridded urban space, integrating multi-source spatial attributes such as load density, land cost, and developability as state inputs. A multi-objective reward function is designed to guide the policy learning process. A simulated urban environment is built for experimentation, and the Advantage Actor-Critic (A2C) algorithm is used for model training and evaluation. The results demonstrate that the proposed method outperforms traditional heuristic approaches in terms of service radius, siting cost, and convergence efficiency, exhibiting strong scalability and practical value. This study provides a novel modeling approach and data-driven optimization tool for smart grid infrastructure planning.