Machine Learning Approach for Identification of Objective Function in Production Scheduling Problems

Yoki Matsuoka, Tatsushi Nishi, Kevin Tierney

Abstract: Systems optimization techniques have become increasingly important in recent years. Experience and knowledge from human experts play a critical role in designing optimization tools for practical uses. These system's evaluation criteria should be selected to accurately reflect the intention of the human operators. In this paper, we propose a machine learning approach for the estimation of objective functions for production scheduling problems. We propose a method to identify the objective function of a problem consisting of the weighted sum of the completion time, the sum of the tardiness, the weighted number of tardy jobs, the maximum tardiness or the sum of setup costs. We consider a supervised learning scenario for predicting an objective function and evaluate several techniques, including a three layer neural network, random forest, and k-neighborhood method. We further investigate feature extraction methods to achieve higher identification accuracy. The effectiveness of the proposed method is verified by comparing the results with methods based on a simplified method that does not use machine learning.