Optimal identification of self-reunion multiple regression (SRMR) model based on regression function for short-term load forecasting

This thesis applies a self-reunion multiple regression (SRMR) model in short-term load forecasting (STLF) and obtains very accurate and steadfast results. This thesis first uses cluster analysis to categorize historical data. Data with similar features will be put in one category. After that, select one group of multiple regression variables in different categories, which serves as the basis for the load forecasting. Then, determine each selected multiple regression variables' regression function for the predicted load by taking the regression function as the base for the forecasting model and using the least-square error. Finally, with the linear programming, find the reunion coefficient corresponding to each regression function. The SRMR model obtained through the fore-going steps is tested by the actual Taiwan load data. Results prove that the average forecast absolute error sought by the model is about 1%, better than the error by the traditional methods.