A hybrid genetic algorithm based on reinforcement learning for the energy-aware production scheduling in the photovoltaic glass industry

Recently, the growing solar energy capacity has played a significant role in developing a clean energy supply system in China. However, the resulting rapid expansion of photovoltaic component (e.g., glass) manufacturing intensifies the energy demand in the locality of the plant. Therefore, this paper considers the energy-aware production scheduling of a deep-processing line in the photovoltaic glass plant, whose layout is a hybrid flow shop with batch and non-batch machines. Firstly, we establish a mixed integer programming model with the minimization of the energy consumption and the penalty for excess of the due date. Then, we propose a hybrid genetic algorithm (GA) based on reinforcement learning to solve the problem. Specifically, the expected Sarsa is used to extract critical knowledge about algorithmic parameters during the population evolution to guide the exploration of the GA. Finally, we conduct extensive numerical experiments to validate the effectiveness of the proposed algorithm by comparing it with a commercial optimization solver and other metaheuristics. The numerical results show that the average gap between the solver and the proposed algorithm is around 4% in small-sized instances. Compared with the heuristic used in the plant, the improvements of this paper are about 16%similar to 18% and 17%similar to 21% in practical-sized instances for the delay penalty and energy consumption objectives, respectively. In addition, the computational results provide managerial insights for managers in further pursuing energy efficiency from higher-level decision-making, e.g., planning over multiple periods from a tactical perspective, and changing production line configurations and introducing new processing techniques from a strategic perspective.