Deep deterministic policy gradient-based symmetry-aware strategy for intelligent wiper system of PV panel

Abstract

Photovoltaic (PV) panels are vital renewable energy sources that convert solar radiation into electricity. However, dust and snow accumulation significantly reduce their efficiency, particularly in arid or cold regions. Existing cleaning systems—whether manual, semi-automated, or robotic—often consume excessive water or fail to adapt to changing environmental conditions. Moreover, current reinforcement learning (RL) approaches such as Q-learning and Deep Q-Network (DQN) lack the flexibility required for real-time optimization. This study proposes a Deep Deterministic Policy Gradient (DDPG)-based intelligent wiper system capable of dynamically adjusting air-blowing and brushing speeds according to debris levels, completely eliminating water use. Simulations were conducted under diverse environmental conditions, including variable dust accumulation rates and solar irradiance levels, to assess system performance. Results show that the DDPG-based model achieved a 35% improvement in panel efficiency, 97% dust cleaning effectiveness, and the lowest mean squared error (0.05) among all tested methods. Operating within a flexible wiper speed range of 1.0–2.5 m/s, it also yielded the greatest reduction in cleaning frequency. The proposed system demonstrates superior adaptability, sustainability, and efficiency, providing a practical, water-free intelligent cleaning solution for maximizing the long-term performance of photovoltaic panels.