Optimal Scheduling of Hydro‐Wind‐Photovoltaic Complementary System Based on Reinforcement Learning‐Proximal Policy Optimization

ABSTRACT The power output of wind power and photovoltaic (PV) features strong stochastic and intermittent characteristics, which increases the difficulty of economic scheduling for the power grid. The hydropower has the advantage of controllable output and fast regulation, which can smooth out the fluctuation of new energy output. By constructing a hydro‐wind‐photovoltaic complementary system (HWPCS), the economy and flexibility of grid scheduling can be improved. To solve the problem of optimal scheduling for the HWPCPS, an optimal scheduling method based on the reinforcement learning‐proximal policy optimization (RL‐PPO) algorithm for the HWPCS is proposed. First, with the objectives of maximizing the economic benefits and minimizing the power fluctuation of power generation, an optimal scheduling model of the HWPCS including multi‐constraints is constructed. Subsequently, by designing the state, action and reward of the RL‐PPO applicable to optimal scheduling, the solution process of the optimal scheduling model for the HWPCS based on the RL‐PPO is designed. Finally, the simulation case verifies that the proposed method can improve the economic benefit of the HWPCS and reduce the power fluctuation caused by the stochastic output of the new energy sources. Compared with deep deterministic policy gradient (DDPG), the twin delayed deep deterministic policy gradient (TD3), and the scenario‐based stochastic optimization (SBSO) algorithms, the RL‐PPO algorithm significantly reduces power fluctuations by 56.44%, 68.59%, and 70.24%, respectively, while maintaining good economic benefits.