Prediction Model of Power Customer Complaints Based on PCA-BP Neural Network

Abstract

In this paper, we constructed a customer complaint prediction model by integrating the following methods: time difference processing, the PCA (Principal Component Analysis), Multiple BP (Back Propagation) neural network calculation, predicted value set processing, event impact index algorithm, etc. The model used the time difference processing and principal component analysis to obtain the optimal solution of the independent variable, used the PCA-BP neural network to fit the function relation between the independent variable and dependent variable, and eliminates the randomness of the neural network algorithm after several accumulative calculations. At the same time, the model established event impact index as model factor to optimize the model's ability to deal with special events. This article realized the effective and accurate prediction of the power customer complaints through the application of the model. Furthermore the model can be applied to other power markets and other areas of forecasting.