APPLICATION OF A HYBRID ARTIFICIAL NEURAL NETWORKS MODEL TO A SCHEDULING POLICY SYSTEM

Abstract

There is a growing trend of application of Artificial Intelligence (AI) to engineering problems. Artificial Neural Networks (ANN) are one of the tools, which has a very simple and easy application in engineering. The training of ANN is done with trajectory dependent algorithms which normally leads to convergence to local minima and limits its application to complex engineering problems. In this paper, a hybrid methodology is proposed where traditional training algorithm in Feedforward Back Error Propagation (BEP) ANN is replaced with Genetic Algorithm (GA) to optimize weights and biases for the ANN and a comprehensive search algorithm to find the optimum number of neurons in the hidden layer(s). The methodology has been applied to an example of a scheduling policy system for flexible manufacturing systems and the results for traditional and hybrid ANN have been compared.