Global Flexible Power Point Tracking Based on Reinforcement Learning for Partially Shaded PV Arrays

The penetration of photovoltaic (PV) systems in modern electricity grids is continuously increasing during the last years. To support the electrical grid by eliminating frequency disturbances and also provide inertia to the power system, the output power of PV systems can be controlled by applying a flexible power point tracking (FPPT) technique, where the PV array output power is continuously regulated to the desired reference value. Since PV modules frequently operate under partial shading conditions (e.g., due to nearby objects in building-integrated PV applications, deposition of dust etc.), the FPPT algorithms designed for operation of the PV system under uniform incident solar irradiance cannot operate efficiently. Therefore, the concept of the global FPPT (GFPPT) has been introduced. In this article, a novel GFPPT algorithm based on machine learning is proposed. By utilizing the proposed Q-learning algorithm, the GFPPT system is able to obtain knowledge over time, in order to achieve faster convergence. The experimental results demonstrated that the proposed GFPPT algorithm converged in significantly less time compared to past-proposed GFPPT methods, while achieving an almost same steady-state tracking error.