A Hybrid Approach Based on Artificial Neural Network (ANN) and Differential Evolution (DE) for Job-Shop Scheduling Problem

Abstract

In this paper, we proposed a new hybrid approach, combining ANN and DE(Differential Evolution), for job-shop scheduling. Job-shop scheduling can be decomposed into a constraint satisfactory part and an optimization part for a specified scheduling objective. For this, an NN and DE-based hybrid scheduling approach is proposed in this paper. First, several specific types of neuron are designed to describe these processing constraints, detecting whether constraints are satisfied and resolving the conflicts by their feedback adjustments. Constructed with these neurons, the constraint neural network (CNN) can generate a feasible solution for the JSSP. CNN here corresponds to the constraint satisfactory part. A gradient search algorithm can be applied to guide CNN operations if an optimal solution needs to be found at a fixed sequence. For sequence optimization, a DE is employed. Through many simulation experiments and practical applica¬tions, it is shown that the approach can be used to model real production scheduling problems and to efficiently find an optimal solution. The hybrid approach is an ideal combination of the constraint analysis and the optimization scheduling method.