STUDIES ON DATA FEATURE AND CORRELATION MINING FOR THE SHORT TERM LOAD FORECASTING

摘要

Input attributes selection and load classification are two key steps that affect the precision of the short-term load forecasting irrespective of the detailed forecasting techniques applied. This paper provides a scheme of data mining for the input variables choice as well as the load patterns classification. The FFT method is adopted in this scheme to filter load data and to analyze the spectral property of long-term hourly-load profiles. Accordingly, the periodical features of the studied power system can be obtained to help the selection of input attributes as well as the model classification. An ANN-based load forecasting scheme is then constructed for the Guang Dong power system. Comparisons with other common adopted input selection prove the effectiveness of this scheme in the short-term load forecasting.