Large-scale scheduling of weaving workshop based on NSGAII and neural network

Objective With the increase of personnel, machines and materials, the scheduling scale of weaving workshop increases exponentially. The intelligent scheduling algorithm represented by genetic algorithm is easy to fall into the local optimal solution when solving large-scale scheduling problems, and the process is slow, which is difficult to meet the actual demand. This study aims to combine the advantages of genetic algorithm and neural network to solve the problem of large-scale scheduling in weaving workshop. Method According to the characteristics of large-scale scheduling of weaving workshop, a weaving workshop scheduling model was established to minimize overdue loss, the makespan and the number of variety changes. Then, a weaving workshop scheduling algorithm NSGAII-NN125 based on NSGAII and neural network was proposed to solve the large-scale scheduling problem of weaving workshop, which consists of a scheduling module and a multi-objective optimization module. Finally, the optimization module was adopted to find the best the scheduling module according to the merits and demerits of the generation scheme, leading to the scheduling module with high quality, fast speed and reusable. Results Comparing the objectives of minimizing overdue loss, the makespan and the number of variety changes, NSGAII-NN125 offered stable performance in a series of weaving workshop scheduling, especially in large-scale scheduling with more than 300 looms and more than 2 000 weaver's beams(Tab. 3). The optimization does not fall into the trend of local optimal solution, and the solution quality is outstanding. Compared with the optimization time, NSGAII-NN125 needed to take longer time to calculate and update the eigenvalues of the neural network. The scheduling speed of NSGAII-NN125 was about 0.67 weaver's beams per second. The NN125 model set was optimized by NSGAII-NN125 according to the scheduling requirements of a weaving workshop which can be used for scheduling similar requirements. Compared with the scheduling objectives, it can be seen that the scheduling quality of the optimized NN125 model set is only slightly weaker than that of NSGAII-NN125, and the time consumption is greatly reduced because the long optimization process is eliminated. The scheduling speed is increased to 50 weaver's beams per second, which has good practical value(Tab. 4). Conclusion The NSGAII-NN125 scheduling algorithm was divided into scheduling module and optimization module in structure. The scheduling and optimization were decoupled, so that the search space of genetic algorithm was limited to a fixed number of parameters in the neural network model, which solves the problem that the GA is easy to fall into the local optimal solution or even scheduling failure due to the large solution space in the large-scale scheduling. More importantly, NSGAII-NN125 outputs the optimal NN125 model set after solving a certain problem. The network model set can be reused to avoid repeated optimization of similar problems and improve the actual scheduling speed, which has good practical value. © 2023 China Textile Engineering Society. All rights reserved.