A Reinforcement-Learning-Enhanced Knowledge-Guided Genetic Algorithm for Flexible Job-Shop Scheduling Problems With Lot Streaming

A flexible job-shop scheduling problem with lot streaming (FJSP-LS) is an extension of flexible job-shop scheduling problems (FJSPs). In it, jobs are allowed to be split into multiple sublots for separate/concurrent processing and transportation, thereby improving job-shop productivity. However, it faces a drastically enlarged solution space due to such sublot splitting, which makes it challenging to find its optimal solution. In this study, a reinforcement-learning-enhanced knowledge-guided genetic algorithm (RKGA) is proposed to solve it. In particular, we design a knowledge-guided strategy that extracts sublot features from sublot schemes (SSs) of an elite solution set as knowledge. This knowledge is then used to guide the SS mutation and state-space generation of the environment in reinforcement learning (RL). Moreover, a method combining perturbation operations and RL is designed to help the algorithm escape from local optima. Extensive computational experiments have been carried out, and the results validate the superiority of RKGA over the state-of-the-art algorithms in solving FJSP-LS.