Influence of GA-BP Artificial Neural Network Based on PCA Dimension Reduction in Yarn Tenacity Prediction

Abstract

The fiber characteristics directly affecting the yarn tensile properties are analyzed and the features of GA-BP artificial neural network are presented. Based on the interrelationship among the fiber characteristics, which could led to a worse predicting result, Principal component analysis (PCA) is adopted to solve this problem. 12 characteristics of cotton fiber tested by HVI or AFIS and yarn process parameters, such as combing, degree of twist, yarn count, was preprocessed by this method and 10 independent comprehensive indexes are induced and substituted into GA-BP artificial neural network as input factors. For further comparison, the 10 dimension reduced indexes and the 15 initial variables are respectively introduced into the GA-BP and BP artificial neural network to develop 4 prediction models. By comparing the accuracy in predicting yarn tenacity, it is concluded that GA-BP model has higher accuracy than BP model, and the dimension reduced indexes based on PCA would decrease the accuracy in prediction instead. So, blindly using PCA method for reducing the complex correlation among input variables was not helpful for the prediction accuracy.