Abstract: For power suppliers, an important task is to accurately predict the short-term load. Thus many papers have introduced different kinds of artificial intelligent models to improve the prediction accuracy. In recent years, Random Forest Regression (RFR) and Support Vector Machine (SVM) are widely used for this purpose. However, they can not perform well when the sample data set is too noisy or with too few pattern feature. It is usually difficult to tell whether a regression algorithm can accurately predict the future load from the historical data set before trials. Here we demonstrate a method which estimates the similarity between time series by Dynamic Time Warping (DTW) combined with Fast Fourier Transform (FFT). Results show this is a simple and fast method to filter the raw large electrical load data set and improve the learning result before looping through all learning processes.
Keywords: Load forecast; Dynamic Time Warping (DTW); Fast Fourier Transform (FFT); random forest; Support Vector Machine (SVM)