Hyper parameter optimization in genetic algorithm with reinforcement learning for production scheduling with lot size

Production scheduling with lot-size involves the process of arranging sequence of products and optimizing number of products to be produced in each batch. The variation in sequence and batch size of production could affect the machine utilization, overall equipment efficiency (OEE), work in process (WIP), and on-time delivery (OTD). The Genetic Algorithm has been used for solving production scheduling with lot size, but it could trap in the local optimal. We propose the method of Reinforcement Learning (RL) and Tabu List to improve GA searching performance and called GARL. RL is used for dynamically adjusting the hyper-parameters. Moreover, the perturbation process is also included to assists GARL to jump out of local optimum. There are three test data sets and three sizes: small, medium, large size. Our results and findings demonstrate that GARL finds better solutions in both of the constraints which included soft and hard than GA even though it takes up a bit more time than GA. The performance of this proposed technique has strengthened the efficiency and minimized the cost of the production process.