基于改进约束强化学习的风光水火储多能互补日内调度方法OA

The large-scale clean energy base is a critical component of China's new power system.Their inherent characteristics—strong uncertainty,low flexibility,and numerous controllable resources—pose dual challenges of timeliness and complexity for intra-day dispatching.This paper proposes an intra-day scheduling method for wind-solar-hydro-thermal-storage multi-energy complementarity based on improved constrained reinforcement learning.First,an intra-day optimal scheduling model for power system is established,incorporating the coordination of wind,solar,hydro,thermal,storage,and DC transmission,and reformulated as a constrained Markov decision process.Second,a policy network architecture embedded with temporal memory units is designed to capture the temporal dependencies of system states,while self-supervised auxiliary tasks are introduced into the value network to enhance feature representation capability.Third,the Lagrangian relaxation method is employed to derive the optimal scheduling policy under safety constraints,and an action correction module based on expert experience is incorporated to jointly ensure satisfaction of both single-step constraints and long-term coupling constraints.Finally,the effectiveness of the proposed method is validated using a modified IEEE 39-bus system and an actual power grid.