The optimization of attribute selection in decision tree-based production control systems

This study develops a learning-based production control system (PCS) to support a manufacturing system to make on-line decisions that are robust in the face of various production requirements. Selecting essential system attributes (or features) based on various production requirements to construct PCS knowledge bases is a critical issue because of the existence of a large amount of shop floor information in a manufacturing system. However, a classical decision tree (DT) learning approach to construct dynamic dispatching knowledge bases does not consider the optimal subset of system attributes in the problem domain. To resolve this problem, this study develops a hybrid genetic algorithm/decision tree (GA/DT) approach for DT-based PCS. The hybrid GA/DT approach is used to simultaneously evolve an optimal subset of system attributes and determine learning parameters of the DT from a large set of candidate manufacturing system attributes according to various performance measures. For a given feature subset and learning parameters of a DT decoded by a GA, a DT was applied to evaluate the fitness in the GA process and to generate the PCS knowledge base. The results demonstrate that the proposed GA/DT-based PCS has, according to various performance criteria, a better long term system performance than those obtained with classical DT-based PCS and the heuristic individual dispatching rules, according to various performance criteria.