Optimal Scheduling of Wind-Solar-Storage Distribution Networks Based on Improved PPO Algorithm

Reinforcement learning (RL) can effectively address the uncertainties of generation and load, and has been applied in the optimal scheduling of power distribution networks with wind, PV, and energy storage. However, classical RL algorithms have drawbacks including performance instability and ineffective exploration-exploitation balance. To address this, this paper proposes an improved Proximal Policy Optimization (PPO) algorithm with dual value networks. A dynamic clipping mechanism is designed to constrain the differences between the old and new networks, effectively balancing exploration and exploitation. Taking the IEEE 33-bus system with wind, photovoltaic (PV), and energy storage as a case study, the proposed improved PPO algorithm is compared with Actor-Critic (A2C), Deep Deterministic Policy Gradient (DDPG), classical PPO, and Dynamic PPO algorithms. The results demonstrate that the improved PPO algorithm achieves optimal performance in terms of training stability, convergence, and the superiority of scheduling results.