Machine Learning and Inverse Optimization Approach for Model Identification of Scheduling Problems in Chemical Batch Plants

Scheduling problems are widely used in recent production systems. In order to create an appropriate modeling of a production scheduling problem more effectively, it is necessary to build a mathematical modeling technique that automatically generates an appropriate schedule instead of an actual human operator. This paper addresses two types of model estimation methods for weighting factors in the multi-objective scheduling problems from input-output data. The one is a machine learning-based method, and the other one is the parameter estimation method based on an inverse optimization. These methods are applied to multi-objectives parallel machine scheduling problems. The accuracy of the proposed machine learning and inverse optimization methods is evaluated. A surrogate model that learns input-output data is proposed to reduce the computational efforts. Computational results show the effectiveness of the proposed method for weighting factors in the objective function from the optimal solutions.