Leveraging reinforcement learning technique to evaluate and predict long-term performance degradation patterns in bifacial photovoltaic modules

Accurate prediction of long-term degradation and energy performance is critical for improving the reliability and lifetime management of bifacial photovoltaic systems. This study presents a unified evaluation framework that compares conventional analytical methods, machine-learning models, and a reinforcement learning based strategy for long-term degradation assessment of bifacial photovoltaic (PV) modules. A consistent multi-year synthetic dataset representing a 50-module bifacial PV system is used to ensure fair and transparent comparison across six approaches. Conventional baseline, current–voltage (I–V) curve, and performance-ratio (PR) models exhibit steadily increasing degradation, reaching 5.6%, 4.65%, and 3.69%, respectively, by the final year. Data-driven models improve prediction accuracy, with random forest (RF) and long short term memory (LSTM) limiting degradation to 3.26% and 3.14%. In contrast, the proposed reinforcement learning approach demonstrates a clear advantage by adaptively optimizing operational decisions, restricting cumulative degradation to only 0.25% in the final year while maintaining stable energy output across all years. Although the reinforcement learning method requires slightly higher computation time due to iterative policy updates, it produces smoother degradation trajectories and superior long-term performance compared to all benchmark methods. Overall, the results confirm that reinforcement learning effectively extends bifacial PV module lifetime while enhancing long-term reliability.