Study and Application of Combination Prediction Model of Principal Component Analysis and Bp Neural Network

When BP neural network is widely used in many fields, this method shows a lot of shortcomings. In this article, the principle component analysis and BP neural network are combined together to establish a combined prediction model. Firstly, we should use principle component analysis to reduce the dimension of the variables and eliminate the co linearity among the variables. According to the selected principle components, BP neural network model will be built. By comparing with the result of single BP neural network model, the fitting degree (R=0.8528) of the combined model is better than that (R=0.7854) of the single one. After comparing the average differences of two models, we have found that the prediction ability (MSE=0.35) of the combined model is better than that (MSE=0.35) of the single model, which shows that the multi co linearity of the resolution factors. It can reduce the data dimension, optimize the structure of neural network, and quicken the speed of the network training and study. The function of the established combined model is good, which can be promoted in the prediction field.