Production Planning for Pineapple Canned with Reinforcement Learning

บทคัดย่อ

Pineapple is a significant economic crop in Thailand, and the pineapple processing industry is crucial for farmers, manufacturers, and customers. Production planning is challenging due to the complexity of customer demands and the increasing uncertainty of fresh pineapple yields. The researcher studied and experimented with algorithms for production planning for canned pineapple using reinforcement learning. The objective was to optimize production planning by finding the best values for just-in-time scheduling through reinforcement learning, based on the Markov Decision Process (MDP) used for sequential decision-making. After analyzing the problem and recognizing that production planning has such characteristics, the researcher proceeded as follows: 1) Designed a dataset from case study data and defined the objective function. 2) Developed a reinforcement learning model using the Advanced Actor Critic (A2C) algorithm to create the production plan for the case study. 3) Tuned the model's parameters, trained the model, and tested it. 4) Evaluated the model and found that the reward or the defined objective function increased by at least 40% compared to the initial model. Additionally, the sellable products' readiness improved by 120%, and the discrepancy between expected and actual returns decreased by 19% compared to the initial machine learning model.