Data-Driven Coordinated Control of AVR and PSS in Power Systems: A Deep Reinforcement Learning Method

Y In this paper, a strategy based on deep reinforcement learning (DRL) as an intelligent coordinator for power system stabilizer (PSS) and automatic voltage regulator (AVR) in a two-are power grid is proposed. The proposed coordinator is developed to provide accurate online modification of the gains appearing in the structure of PSS and AVR which avoids unfavorable interactions between PSS and AVR under significant changes in the working point and thereby guaranteeing the stability of the power grid. A Markov decision manner is used to formulate the DRL problem and it is solved through a deep deterministic policy gradient approach with an actor-critic framework. Since the intelligent coordinator relies on the expert's science, some scaling coefficients are added to the coordinator body to achieve optimal performance. To confirm the effectiveness of the presented DRL approach, the design is conducted on Kundur's power grid. Simulations illustrate that the proposed DRL-based control can confirm the stability of the system and attain desired dynamic responses.