Real-time Economic Dispatch of Thermal-Wind-Battery Hybrid Systems based on Deep Reinforcement Learning

In the context of large-scale renewable power penetration and extensive deployment of energy storage devices, real-time economic dispatch is of great significance to guarantee reliable and cost-effective power system operations. Traditional dispatch methods may have difficulty in rapidly responding to the fluctuation and uncertainty of both the power supply and demand sides. Therefore, a deep reinforcement learning-based method is proposed in this paper to formulate real-time economic dispatch strategies according to the ultra short-time rolling forecasts of renewable power and load demand and the state-of-charge of battery banks. Numerical experiment results on a thermal-wind-battery hybrid system exemplify the effectiveness of the proposed method and show better performance compared to benchmarks.