A Novel Reinforcement Learning Algorithm-Based Control Strategy for Grid-Configured Inverters

To address the power oscillation problem due to the introduction of inertia and damping, this paper proposes a new deep reinforcement learning algorithm based on the SD3 (Softmax Deep Double Deterministic policy gradients) algorithm for grid configuration inverter control strategy to compensate for the loss of inertia and damping in the grid. Virtual synchronous generator control, as a typical grid configuration technique, inevitably brings stability problems while providing inertia and damping support to the grid. In this paper, we first analyze the nonlinear relationship between inertia and angular velocity, and find the key parameters to maintain the power stability; then, we migrate the deep reinforcement learning strategy, and design the control strategy applicable to the virtual synchronous generator; finally, through the adaption of the key parameters, we combine the control strategy with the grid-connected inverter, and solve the problem of the excessive grid-connected power oscillation of the inverter. The effectiveness and accuracy of the control method compared with other algorithms are verified by building a simulation model in MATLAB/Simulink, which realizes the purpose of reducing power oscillation.