A Graph Embedding Algorithm Based on Reinforcement Learning for Solving Fuzzy Multi-Objective Flexible Job Shop Scheduling Problem

This study delves into the multi-objective flexible job-shop scheduling problem with fuzzy time (MOFFJSP), addressing complex and dynamic production environments, with the goal of minimizing the maximum completion time and total machine workload (TMW). A graph embedding algorithm based on reinforcement learning (GEARL) is proposed in this article, consisting of four modules: population initialization, fuzzy disjunctive graph, policy model, and solution set optimization processing. Diverse initial populations are constructed using different rules tailored to the problem. A fuzzy disjunctive graph is designed to transform individual data information into graph information. Graph embedding technology is utilized to extract individual feature information and generate corresponding optimization strategies through the policy model. Solution set optimization processing involves performing non-dominated sorting on the entire population to filter out advantageous individuals to guide subsequent optimization directions. Experimental results demonstrate that the GEARL algorithm exhibits significant advantages in solving the MOFFJSP.