Monthly Power Load Predicting by WT and LS-SVM

Fast wavelet transformation can decrease the noise and the correlation among the monthly power load information. A new machine learning method-least square support vector machine (LS-SVM), based on the fast wavelet transformation (WT), was used to build the model to forecast monthly power load. Definition and application of the fast WT and the LS-SVM were introduced. The sym4 wavelet basis was selected as the wavelet function and the WT level was 3. The denoised monthly power load by the fast WT was compared with the original power load. Mean relative error (MRE) and root square mean error (RSME) of the direct LS-SVM prediction of the power load was 6.0045 percent and 1219 million kilowatt-hour (MKWH) respectively. MRE and RSME of the WT-LS-SVM was 3.88 percent and 845 MKWH respectively. Excellent forecasting accuracy of the WT-LS-SVM can provide the long-term power load forecasting an effective ways.