Short-Term Load Forecasting Based on Fuzzy Clustering Analysis Similar Days

Abstract

As to the short-term electric power load forecasting, its accuracy is affected by many uncertain influencing factors. To improve the forecasting accuracy, a novel method using Similar Days based on fuzzy clustering analysis is proposed in this chapter. Firstly it categorizes the weather factors as temperature, air pressure, wind speed, overcast day, rainy day, etc., and then together with week type and day type these factors form the influence items. According to the items above, fuzzy rules are applied to establish the mapping table to get the factors quantized. Next, the cluster technology is utilized to classify the content in the mapping table, and the similar days are chosen based on the clustering level, which is to reduce the numbers of samples and accelerate the speed of selection. Secondly, aiming to eliminating non-gaussian noise contained in the similar days’ power load, lifting wavelet transform is adopted to extract the low sequence components. Finally a Least Squares Support Vector Machine (LS-SVM), which is optimized by particle swarm optimization algorithm, is designed to predict the low-frequency part while mean square weighted method is used to predict the high-frequency part. The simulation results show that this fuzzy clustering similar days method is effective.