Abstract:

Inversion model of traditional worsted spinning process parameters is defective because of unsatisfactory convergence and stability, poor generalization performance and low inversion accuracy. Besides, standard genetic algorithm (SGA) shows premature convergence in the application of complicated optimization problem. Therefore, the paper put forward an inversion model of mixed population genetic-artificial neural network (MPG-ANN) on the basis of BP neural network. First of all, the author established the prediction model by optimizing the weights and threshold of BP neural network through mixed population genetic algorithm. On this basis, the author created the inversion model of mixed population genetic algorithm according to the CV value of yarn and used it for the inversion of process parameters during worsted spinning production. Next, by taking lots of test data of field process in spinning workshops as the object, the author conducted inversion verification with the wool top oil content and spinning draft multiple of process parameters. The results show that the inversion accuracy of MPG-ANN model reaches 97% and increases by 4% relatively to SGA-ANN model. In addition, the fluctuating margin of the inversion results of MPG-ANN model reduces by 6.28% at most relatively to SGA-ANN model. Finally, the author predicted the CV value of yarn on the basis of inversion results. The average prediction accuracy increases by 4.25% relatively to SGA-ANN model, showing MPG-ANN model is superior to SGA-ANN model in terms of generalization performance.