The Application of Two New Integrated Models in Short-Term Load Forecast

Based on four general forecasting models (SVM model, BP neural network model, wavelet regression model and similar date model), two new models (integrated model I and integrated model II) are proposed in this paper. In the process of determining models and parameters, the virtual forecast conception is adopted. And a series of improvements on the aspects of historical data, temperature factor, holiday factor, economic growth, etc are made. Finally a global forecast competition which was held by EUNITE Network on August 1st, 2001 is taken as an example, showing the average daily error and maximum daily error in the forecasting of this two integrated models have been improved obviously with respect to the four simple models. So the two integrated models are proved to have has an important practical value.