Novel adaptive stability enhancement strategy for power systems based on deep reinforcement learning

As the access rate of wind energy in a power system has significantly increased, stabilizing the power system has become challenging. Among these challenges, low-frequency oscillation is one of the most harmful problems, effectively resolved by adding a damping controller according to the relevant properties of the low-frequency oscillation. However, the controller often fails to adapt to the constantly changing wind energy system owing to the lack of a targeted dynamic change strategy. Thus, to address this issue, an adaptive stabilization strategy that uses a static var compensator with an additional damping controller structure is proposed. Specifically, the entire power system is equivalently represented as a generalized regression neural network, with a deep reinforcement learning algorithm called soft actor-critic introduced to train the agent based on the generalized regression neural network model. After the training process, the agent can provide additional efficient static var compensator damping controller parameters under different operating conditions, vastly improving the system stability. Simulation results verify the improved performance using the proposed strategy compared to other optimization methods, regardless of whether the low-frequency oscillations were suppressed in the time or frequency domains.