Modeling of flow shop scheduling with effective training algorithms-based neural networks

Abstract

This paper deals with the performance comparison of three most effective neural network backpropagation training algorithms such as gradient descent, Boyden, Fletcher, Goldfarb and Shanno (BFGS) based Quasi-Newton (Q-N) and Levenberg-Marquardt (L-M) algorithms. The training of the neural network is carried out based on random datasets considering optimal job sequences of the permutation flow shop problems. In the present investigation, a goal of 0.001 of MSE or 3000 of epochs is set as a goal of learning. The overfitting and overtraining are not allowed during model building to avoid poor generalization ability. The performance of different learning techniques is reported in terms of both solution quality and computational times. The computational results demonstrate that the L-M performs best among the three algorithms with respect to both MSE and R2. However, the gradient descent algorithm is the fastest among them.