Learning Based Memetic Algorithm for the Monocrystalline Silicon Production Scheduling Problem

Abstract

In recent years, with the rapid development of the global semiconductor industry, the importance of monocrystalline silicon is increasingly important. This paper addresses a novel monocrystalline silicon production (MSP) scheduling problem. We analyze the characteristics of the MSP, and the MSP is molded as an unrelated parallel machine scheduling problem with maintenance cycle and machine setup time. To solve the MSP problem, a learning based memetic algorithm (LMA) is proposed. In the LMA, first, a high-quality initial population is constructed according to problem-specific knowledge, as well as an initial probability distribution model is established to accumulate valuable information about superior individuals. Second, to improve the ability of global exploration, an adaptive update mechanism based on the probability distribution model is developed, and a sampling method that keeps excellent patterns was designed to generate a new population. Third, a local search strategy based on variable neighborhood descent (VND) is designed to enhance the local exploitation ability. In the VND component, to enrich the search behavior, four self-adaptive neighborhood operators are devised. Moreover, we applied the simulated annealing (SA) as acceptance criteria of solutions to avoid the algorithm falling into a local optimum. Finally, simulation experiments based on some real-world instances from an MSP process demonstrate the effectiveness of the proposed LMA in solving the MSP scheduling problem.