Towards sustainable closed-loop photovoltaic supply chains: A hybrid framework integrating reinforcement learning and mathematical models

Abstract

In light of the deleterious consequences of excessive fossil fuel consumption, the transition to renewable energy sources, particularly solar energy, is inevitable. Consequently, designing an efficient supply chain that enhances recycling management and achieves sustainability goals is imperative. This study unveils an innovative approach for the sustainable management of photovoltaic system supply chains, employing a two-stage methodology based on reinforcement learning and mathematical modeling. Given technological advances, a dynamic environment, and uncertainties in selecting solar panels, the first phase uses reinforcement learning to identify the panel with the highest efficiency. The second stage focuses on designing a closed-loop supply chain network to minimize economic costs, mitigates environmental impacts, and incorporate social sustainability into decision-making. Likewise, time series models for forecasting photovoltaic energy demand enhance the model's accuracy and efficiency. The proposed approach includes a comprehensive case study in Iran to demonstrate its real-world applicability. Key results indicate that the crystalline silicon panel emerges as the optimal choice through reinforcement learning. The demand forecasting model reveals an upward trend, and multi-objective optimization produces balanced network configurations that support trade-offs among economic, environmental, and social.