Development of deep reinforcement learning for maximum power point tracking of photovoltaic systems

Abstract

The use of renewable energy systems, specifically photovoltaic (PV) systems (PVs) that convert solar energy into electricity, has become a popular solution to address global environmental concerns by reducing the utilization of non-renewable energy sources, which contribute to pollution. Efforts to increase the power transfer effectiveness of PV systems include the advancement of controllers for maximizing power point tracking (MPPT). These controllers guarantee optimal system operation at the maximum power point (MPP) in diverse environmental conditions. The paper proposes an improved deep reinforcement learning (DRL) method, namely deep deterministic policy gradient (DDPG), to capture the MPP in PV systems, particularly when dealing with partial shading conditions (PSCs). Unlike reinforcement learning methods that only work with discrete state and action spaces, the proposed DDPG method can handle continuous action state spaces. Feasibility analysis is conducted using MATLAB/Simulink simulations, and the findings demonstrate the efficiency and superior performance of the suggested solutions, highlighting their potential for future use.