A Novel Hopfield Neural Network Approach to Job-Shop Scheduling Problems

This paper proposes a novel method based on Hopfield neural networks (HNNs) for solving job-shop scheduling problems (JSPs). The JSP constraints are analyzed and their permutation matrix express is developed. A new calculation energy function is also proposed, which includes all JSP constraints. A novel Hopfield neural network for such JSP problems is constructed and the effect of its weights for the scheduling problems is investigated. To avoid the HNN converging at local minimum points and to generate some non-feasible scheduling solutions for JSP, this paper applies a simulated annealing algorithm to the HNN. Compared with traditional methods, this novel method helps the HNN to converge at the minimum volume 0, which can guarantee that the steady outputs of neural networks will be a feasible solution for a specific job-shop scheduling problem.