Research into Power Load Forecasting Based on Strong Regression Wavelet Neural Network with Variable Basis Functions

摘要

The Wavelet Neural Network (WNN) is widely used in power load forecasting. In view that the traditional WNN easily falls into the local minimum and has unstable forecast results, a new power load forecasting model of combining the AdaBoost algorithm with WNN was put forward to improve the forecasting accuracy and generalization ability. Firstly, the method performed the pre-treatment for the historical power load data and initialized the distribution weights of test data. Secondly, it selected different wavelet basis functions randomly to construct weak predictors of WNN, and trained the sample data repeatedly. At last, it made more weak predictors of WNN to form a new strong predictor by AdaBoost algorithm for regression forecasting. A simulation experiment for the dataset of Individual Household Electric Power Consumption in University of California Irvine (UCI) was carried out. The results show that this method has reduced the average error value by more than 66.5% compared to the traditional WNN, and has improved the forecasting accuracy of neural network. This method provides references for the WNN forecasting.