Power Short-term Load Forecasting Method Based On Data Mining

Short-term power load forecasting(SPLF) has strong influence on the operation, control and plan ofpower systems. The accurate forecasting provides maintaining security and stability to the operation ofpower systems. Based on various data mining technologies, The pre-process of historical load data, processof weather condition, establishment of forecasting model and its input parameters mining, are studied inthis paper. The studies in this paper are aimed at laying a solid foundation for high accuracy SPLF softwaredevelopment.The measures distance by kernel functions instead of the complicated Euclidean distance and thiskernel based distance is used as dissimilarity function of target clustering formula which can reduce thecalculation complexity. In addition by subtraction clustering algorithm to initialize clustering center of thefuzzy c-means clustering and clustering number. After the clustering, a radial basis function neural networkbased identification model for load data is proposed, and the bad data is modified, The experiment resultsshow that the proposed data processing model has good effect.As the similar days is concerned, A method is proposed based on improved similarity of fuzzyclustering algorithm. First, the weights of the meteorological factors are obtained through path analysis. Animproved similarity is constructed integrating weighted similarity coefficients and weighted distancecoefficient. Second, the history day samples are divided into several categories by making fuzzy similaritymatrix.Finally, based on the former analysis, SVM model is used for the prediction model of short-term loadof normal days. In our experiment, the data stems from the real electric power load data of foshan city,guangdong province. Compared the proposed model in our paper with SVM power prediction model,results show that our method has better prediction accuracy. And the proposed model is also used topredict the preprocessing data and original data. Experiments show that preprocessing data can improve theprediction accuracy.