Deep reinforcement learning-based robust nonlinear controller for photovoltaic systems

Recently renewable energy such as a photovoltaic (PV) system has been utilized more and more since it is pollution-free and permanent. To maintain the PV system functioning at, or near, the peak power point of the PV panel under different conditions such as fluctuating solar irradiation, temperature, and other factors, maximum power point tracking algorithms are required. In this study, a novel hybrid robust intelligent controller for a photovoltaic system is proposed. Three loops are used for creating the proposed controller, ensuring the controller’s robustness. The first loop’s objective is to locate the photovoltaic system’s highest power spots. In the second loop, a novel fractional-order sliding model observer based on deep reinforcement learning optimization approach is proposed as a result of the design of a reliable controller under the lumped uncertainty in the system. Designing a novel back-stepping fast non-singular terminal fractional-order sliding mode controller is achieved at the final step. This method offers a nice transition response, a small tracking error, with a quick response to changes in solar radiation. Numerical analysis shows that the performance of the photovoltaic system by the proposed controller has been able to increase the efficiency of the system significantly.