Optimal Operation of Port PV Systems Based on Improved Reinforcement Learning

ABSTRACT Introducing new energy sources to create green ports holds great importance in the context of global energy conservation. As an important source of new energy power generation, the optimal operation of photovoltaic (PV) power generation has always been the focus of research. Port PV systems are prone to partial shading due to variable weather conditions and building occlusion. This phenomenon results in PV arrays with multipeaked power curves, thereby increasing the difficulty of maximum power point tracking (MPPT). Moreover, traditional reinforcement learning (RL) is incapable of addressing this problem effectively. Therefore, this paper proposes an improved RL‐MPPT algorithm. To enhance the convergence speed of this method, a Q‐table matching mechanism based on a back propagation neural network (BPNN) is introduced. Additionally, to mitigate the problem of the PV panel aging and the lack of updates in the Q‐table knowledge base, a Q‐table knowledge base time window update mechanism is implemented. Finally, MPPT method is verified by a comprehensive set of simulation experiments. The experimental results indicate that the Q‐table matching mechanism based on BPNN can effectively improve the convergence speed of the RL algorithm, and can effectively overcome the multi‐peak power characteristics in the MPPT control process of the port PV system in complex environmental scenarios, with good control accuracy, dynamic characteristics and robustness.