REINFORCEMENT LEARNING BASED MAXIMUM POWER POINT TRACKING CONTROL OF PARTIALLY SHADED PHOTOVOLTAIC SYSTEM

Under the sun insolation in the daytime, the Maximum Power Point Tracking (MPPT) technique is usually used to achieve the maximum power in the photovoltaic (PV) system and often implemented by the Perturbation and Observation (P&O) method. However, due to the use of fixed step size, the P&O method will generate undesired oscillation around the maximum power point (MPP) and thus reduce the tracking efficiency. Besides, the output power of PV modules highly depends on the environment factors such as irradiance and temperature, especially for a PV array, which is formed by PV modules connected in series and parallel. The partially shaded effect would easily happen in a PV array due to clouds, buildings, trees, etc. Due to the partially shaded effect, the characteristic P-V curve of a PV array may possess multi-peaks, which often results in tracking of a local maximum, not the expected global maximum. To deal with the partially shaded effect, this paper proposes a Reinforcement Learning based MPPT method, which is implemented by Q-learning method. Demonstrated by numerical simulation results, the proposed method indeed can track the global MPP faster and more precisely without oscillation.