Design of reinforcement learning based robust μ-synthesis controller for single phase grid-connected VSI

Grid-connected voltage source inverters have been increasingly employed for a wide range of applications such as power conversion and conditioning, filtering, compensation, etc. The increased employment of the number of inverters due to excessive penetration of Distributed Energy Resources (DER) also causes huge concern to the overall stability of the system in terms of voltage fluctuations caused due to changes in grid impedance. Most of the existing solutions offer complex findings when it comes to practice for weak grids and the presence of non-linear loads. In this paper, a $\mu$-synthesis controller for change in the impedance issue statement has been designed to study a robust solution to the controller problem. The order of the controller has been reduced with a balanced model reduction approach. A novel methodology of tuning the weighting functions of the controller with advanced machine learning-based reinforcement learning has been adapted and performance specifications of the controller have been studied with tuned weighting functions. To investigate the controller efficacy on non-linear loads, the viability of the controller was validated with real-time implementation. The stable operating modes were demonstrated by hardware in loop simulation of the inverter, and the controller along with the non-linear load.