Deep Reinforcement Learning Based MPPT Control for Grid Connected PV System

• Main Authors: Vora, Kunal, Liu, Shichao, Dhulipati, Himavarsha
• Format: Conference Proceeding
• Language: English
• Subjects: Deep reinforcement learning; Energy Storage; Heuristic algorithms; Maximum power point trackers; Maximum power point tracking; Perturb and Observe Algorithm; Photovoltaic System; Q-learning; Renewable energy sources; Simulation; Software packages; Sustainable Energy
• ISSN: 2769-3899
• DOI: 10.1109/ICPS59941.2024.10639977

Maximum power point tracking (MPPT) helps in generating maximum power from PV system at a specified irradiance levels irrespective of changes in the sun's position and cloud cover conditions. From previous studies, it is observed that, conventional methods for MPPT suffers from oscillations around maximum power point and does not adapt to changing environmental conditions of irradiance and temperature. Therefore, new techniques like reinforcement learning is implemented in PV system to overcome aforementioned limitations. In this paper, integration of deep learning and reinforcement learning named deep Q-learning (DQN) is implemented in grid connected PV system. DQN solves the problem of varying environmental conditions by discretizing the state spaces. The proposed method is implemented in MATLAB/ SIMULINK environment. Based on the simulation results, it can be proposed that proposed method is efficient in handling ever changing environmental conditions.