A reinforcement learning approach for MPPT control method of photovoltaic sources

Photovoltaic arrays are the means to convert solar power into electricity, and a significant way to generate renewable and clean energy. To be efficient, a photovoltaic must generate constantly the maximum possible power and under different environmental conditions. Finding the maximum generated power has been a known issue in the industry using methods of classic control theory with very good results. However, those solutions are case-specific resulting to increased set-up effort. This work proposes a universal RLMPPT control method based on a reinforcement learning (RL) method that tracks and adjusts the maximum power point of a photovoltaic source without any prior knowledge. A Markov Decision Process (MDP) model for the Maximum Power Point Tracking (MPPT) photovoltaic process is defined and an RL algorithm is proposed and evaluated on a number of photovoltaic sources. The proposed RLMPPT control method has the advantage of being applicable to different PV sources with minimum set-up time. To evaluate the RLMPPT control method performance, a number of simulations run under different environmental and operating conditions and a comparison with the conventional method of Perturb and Observe (P&O) is performed. Results show quick response and close to optimal behavior without requiring any prior knowledge.