Short-term Power Load Forecasting Based on Density Peak Clustering and Improved LWLR
Short-term power load data are characterized by complexity and uncertainty,which often have an uncontrollable impact on the prediction results of the data. When short-term electricity load data are clustered and analyzed using traditional clustering methods,the prediction results will be biased due to characteristics such as the uncertainty of electricity load. In addition,considering that global regression forecasting methods are unable to use different modeling approaches for different parts of the data in the modeling stage,which limits the problem of adaptive capability for different distribution regions or different subsets of features. In this paper,we adopt the density peak clustering algorithm with K-nearest neighbor and weighted similarity to classify the features of short-term electricity load data,and propose a locally weighted linear regression model using K-nearest neighbor to forecast short-term electricity load.The advantages of this model are that it avoids the influence of Euclidean distance on the selection of cluster class centers,reduces the negative influence of global data on local data,avoids the centralized effect of cluster class division,and improves the generalization ability of the model. By comparing with fuzzy C-mean clustering and traditional global regression prediction methods,the model proposed in this paper is more superior for the prediction of real power data.
density peak clusteringK nearest neighborlocally weighted linear regressionpower load forecastingprediction performance evaluation
万方数据
维普
