Short-Term Load Forecasting for Similar Days Based on Support Vector Machine and Dempster-Shafer Theory

In view of the defects in the load forecasting based on support vector machine (SVM) such as high dimensionality of input data and long training period, an evidence fusion-based load forecasting method, in which the SVM is utilized, for similar days is proposed. During the choosing of similar days the range of average load, the shape of load curve and the difference of temperature are considered, and by means of evidence fusion the similar day, whose load is highly similar to the forecasted day, is obtained and is taken as the training data of SVM while a lot of redundant data are rejected, thus the dimensions of the input is decreased and the forecasting accuracy is improved. Applying this method to short-term load forecasting and comparing the forecasting results with those by standard SVM, it is proved that the forecasting accuracy is evidently improved.