Research on AGC Adaptive Control of New Energy in Power Grid Considering Security Deep Reinforcement Learning

In order to reduce the response time and frequency deviation of Automatic Generation Control (AGC) control for new energy generation, a secure deep reinforcement learning based adaptive control method for grid new energy AGC is proposed. A new energy AGC adaptive control objective function was constructed with the goal of optimizing production costs, including unit cost parameters and power generation constraints, and ensuring system safety through node power balance equations and inequality constraints such as voltage and line power. Aiming at the safety risks of deep reinforcement learning in power grid control, the Safe Soft Actor Critic (SSAC) algorithm is proposed, which transforms the frequency control model into a Constrained Markov Decision Process (CMDP) and strictly incorporates safety constraints such as voltage and power. By defining the state value function and action value function, combined with the Bellman equation iterative optimization strategy, it ensures that the action execution does not exceed the limit, and quantifies the voltage deviation and power limit number with the cost function. Finally, the CMDP problem is solved by updating the expression of the strategy, achieving the coordinated control of safety and economy. The experimental results show that the control response time of the proposed method is only 0.18 to 0.22 seconds, and the frequency deviation rate is stable below 1%, which is significantly better than the comparative method.