Primary Frequency Control of Renewable Energy Based on Deep Reinforcement Learning

Abstract

Proportion of wind power, photovoltaic, and other renewable energy sources in the power system gradually increasing, the capability of system primary frequency regulation has a trend of weakening. To improve this situation, renewable energy sources are required to have the same ability to participate in system frequency regulation as conventional power sources. This paper analyzes the dynamic performance of the interconnected power system composed of wind farms based on doubly-fed induction generators (DFIG), then proposes a control method based on the deep reinforcement learning algorithm-Deep Deterministic Policy Gradient Agents (DDPG), according to the power generation characteristics of wind farms. The reward function, input states, and output actions of the DDPG algorithm are designed in conjunction with the control objectives, thus the DDPG algorithm is effectively applied to the primary frequency regulation optimization scheme to achieve the control goals which is adaptively acquiring the optimal coordination control strategies of the controllers of multiple renewable energy power plant. Numerical simulations on a two- area power system demonstrate that the design scheme can effectively mitigate the regional frequency deviation problem of each area containing in the power system, to maintain the stable operation of the system.