The Smart Worsted Spinning Forecast Model Based on BP Neural Network and Principal Component Analysis

The characteristic of worsted spinning procedures and the BP neural network modelling technology all have been summarily analyzed. Based on the nonlinear and vague relationship depend on fibers' performance and craft parameters, the principal component analysis method are proposed to pretreat the sample data set, which results are the new sample data of the BP neural network. Therefore, the input layer node numbers reduce; the relevancies among every input factor are eliminated simultaneously; the network topology structure is simplified that the network's accuracy and performance are all enhanced greatly. After modelling, the relative mean error percents (MEP) between the predict results and measured value for the yarns' four quality variables, such as Yarn unevenness, strength, extension at break and ends-down rate, reduce to 2.84%, 2.31%, 2.98% and 1.96% respectively compared to the former 4.62%, 3.38%, 4.35% and 3.87%. The correlation coefficients between them for the four quality variables all have the remarkable enhancement.