Due Date Determination in Dynamic Job Shop Scheduling with Artificial Neural Network

In this study, an artificial neural network approach that is thought to produce better results as an alternative to due date determination methods in dynamic job shop scheduling environment is presented and its feasibility is demonstrated. The performance of the neural network model is compared with five different regression models. An event oriented simulation software is developed for the determination of the coefficients of the regression models and for the generation of data to be used in the training of the neural network model. Back-propagation artificial neural network was used as an artificial neural network model and a software was developed. After the regression models were created and the neural network was trained, the simulation software was run for the shortest processing time and earliest due date priority rules for comparison purposes. In order to compare the models, average absolute deviation from the due date, mean square of absolute deviation from the due date, average tardiness, number of tardy jobs, average earliness and number of early jobs were used as performance metrics. As a result of the study, the artificial neural network model was found to be effective in due date determination. Both the shortest processing time first and the earliest due date first priority rules gave good results in terms of several performance metrics. It was observed that the neural network gave better results in the shortest processing time priority rule in general.