Soft Actor-Critic-Based MPPT Control of Solar PV Systems Under Partial Shading Conditions

This paper presents a soft actor-critic (SAC)-based method for solving the solar photovoltaic (PV) Maximum Power Point Tracking (MPPT) control problem under partial shading conditions (PSCs). The MPPT method optimizes the solar PV power and ensures that it constantly operates at its “maximum power point (MPP),” regardless of the dynamics of weather conditions. Traditional MPPT methods, such as the perturb and observe (P&O) method are commonly employed to solve the MPPT control problem. However, they often suffer from a slower convergence rate, significant oscillation near the MPP, drift problems, and frequently fail to track the solar PV global maximum power point (GMPP) in the presence of partial shading. These problems were addressed using the deep Q-network (DQN) method. However, DQN cannot be applied to continuous action spaces. It also uses inefficient experience replay and suffers from Q-value overestimation. Thus, under PSCs and certain environmental conditions, DQN produces fluctuations of power close to the MPP or GMPP, resulting in power loss. To solve the MPPT control problem, a mathematical model based on a Markov Decision Process which allowed the SAC agent to learn the solar PV behavior was developed. Key hyperparameters affecting the SAC algorithm's performance were also investigated. Furthermore, the P&O method were developed for comparison. Simulation results show that the SAC-based MPPT method achieved better tracking accuracy than the DQN method under standard testing conditions, varying irradiance levels, and PSCs. Also, it is shown that both the DQN and SAC methods have superior tracking performance compared to the P&O method under similar environmental conditions tested.