AHOA: Adaptively Hybrid Optimization Algorithm for Flexible Job-shop Scheduling Problem

The Flexible Job-shop Scheduling Problem (FJSP) is a typical scheduling problem in industrial production that is proven to be NPhard. The Genetic Algorithm (GA) is currently one of the most widely used algorithms to address the FJSP task. The major difficulty of using GA to solve FJSP lies in how to set the key hyperparameters and improve the convergence speed. In this paper, we propose a hybrid optimization method based on Reinforcement Learning (RL) called the Adaptively Hybrid Optimization Algorithm (AHOA) to overcome these difficulties. The proposed algorithm first merges GA and improved Variable Neighborhood Search (VNS), which aims to integrate the advantages of global and local search ability into the optimization process. Then the double Q-learning offers crossover and mutation rates according to the feedback from the hybrid algorithm environment. The innovation of this work lies in that our method can adaptively modify the key hyperparameters in the genetic algorithm. Furthermore, the proposed method can avoid the large overestimations of action values in RL. The experiment is evaluated on the most widely studied FJSP instances and compared with some hybrid and self-learning algorithms including dragonfly algorithm (DA), hybrid gray wolf weed algorithm (GIWO), and self-learning GA (SLGA), etc. The results show that the proposed method outperforms the latest related algorithms by more than 12% on average.