Yarn Unevenness Prediction Using Generalized Regression Neural Network

Abstract

This study aimed to propose a method to predict yarn unevenness grounded on the generalized regression neural network and traditional neural network model to further improve the prediction accuracy. The yarn unevenness model was constructed. Under this model, a three-layer neural network, a four-layer neural network, a five-layer neural network, and a generalized regression neural network were designed. Finally, Python was used for training and simulation. The training parameters and the three network models data were made consistent to ensure the comparability of the results. The results showed that using the yarn unevenness model, the average relative error of the four-layer neural network to cut down 0.87% compared with that of the three-layer neural network. Compared with the five-layer neural network, the four-layer neural network performance was not much different, but the running speed was increased by 46.05%. Compared with the four-layer neural network, the average relative error of the generalized regression neural network was reduced by 0.57%, the mean square error was reduced by 0.98%, he root mean square error was reduced by 4.76%, and the running speed was increased by 74.70%.