Improved Particle Swarm Optimization Algorithm Combined with Reinforcement Learning for Solving Flexible Job Shop Scheduling Problem

Particle Swarm Optimization (PSO) is widely used to solve optimization problems. Most existing PSO algorithms only improve on inertia weights, but not in speed, position, and learning factors. Particles cannot find the optimal value more accurately based on their current position. In this paper, an improved particle swarm optimization algorithm combined with reinforcement learning (IPSO_RL) is proposed to solve the flexible job shop scheduling problem (FJSP) with the optimization goal of minimizing the maximum completion time (makespan). At first, in the particle update stage of this algorithm, a Q-learning algorithm is proposed to dynamically adjust inertia weights and acceleration parameters to balance the algorithm's global exploration and local exploitation capabilities, thereby guiding the search direction reasonably. Secondly, a particle position update strategy was redesigned to accelerate the convergence speed and accuracy of the algorithm, to improve search efficiency. In addition, the algorithm introduces an opposition-based learning strategy that can enrich the search direction of the solution space and enhance the algorithm's ability to jump out of local optima. Finally, simulations experiments, and comparisons demonstrate that IPSO_RL can effectively solve the FJSP.