Maintaining the Frequency of AI-based Power System Model using Twin Delayed DDPG(TD3) Implementation

This paper proposes a multi-agent deep reinforcement learning (MA-DRL) method for load frequency control of a renewable energy single-area power system in a continuous action-space domain. This method can non-linearly adapt the control strategies for cooperative LFC control through off-policy learning. Multi-agent twin delayed deep deterministic policy gradient (TD3) is proposed to adjust and refine the control system parameters considering variational load and source behaviour. Implementation of the model requires only local information for each control area to achieve an optimal control state. Comparison between TD3 and DDPG model proves the edge of TD3 model. Simulations and numerical data comparison on a renewable energy single-area power system demonstrate that the proposed model can successfully reduce control errors and stochastic frequency deviations caused by load and renewable power fluctuations.