Parameter Optimization of Multiple Resonant Controller: A Deep Reinforcement Learning Approach

PR controller is widely applied to the control of periodic voltage with zero steady-state error. However, considerable design effort is required to ensure stable operation and robustness to parameter uncertainties especially several resonant controllers paralleled. This paper deals with the synthesis problem of multiple resonant controllers. Taking harmonic generator as an example, a new method using reinforcement learning is presented, by which the parameters of resonant controllers can be automatically tuned. First, the feature of multiple resonant controllers is analyzed under various load conditions. Then, realization of the reinforcement learning approach is discussed in detail, among which security is emphasized. Simulation validates correctness of the proposed method and experimental framework is presented.