Abstract:

In order to further improve the accuracy and training speed of the BP neural network in yarn quality prediction, a four-layer BP neural network with double hidden layers was proposed for cotton yarn quality prediction on the basis of the conventional three-layer BP neural network model of single hidden layer. By constructing the model of the breaking strength of pure cotton yarn and the CV model of yarn levelness, a three-layer BP neural network and a four-layer BP neural network were designed under each model, and the final training and simulation were performed using MatLab. In order to ensure the comparability of the final results, the training parameters of the two network models and the data used are consistent. The experimental results show that under the fracture strength model, the maximum number of training steps in the four-layer network compared to the three-layer network is reduced from 740 to 533, and the relative average error decreases from 9.6% to 7.5%. In the yarn levelness CV value model under the four-layer network, compared with the three-layer network, the maximum number of training steps decreases from 929 to 604, and the relative average error decreases from 10.2% to 8.3%.

Key words: yarn quality prediction, cotton yarn, four-layer BP neural network, MatLab simulation