Research On Load Forecasting Method Of Distribution Network Based On Improved K-means Clustering

With the continuous development of China’s power system and the modernization and complexity of power grid management,accurate and effective power load forecasting has become increasingly important.For example,excessive energy consumption requires people to pay attention to the utilization of electricity resources.The research results of power load forecasting can predict the energy consumption of a region,guide the operation of the power grid,and have great significance in promoting the realization of the national goal of "carbon peaking,carbon neutrality".In existing research on power load forecasting,methods mainly focus on regional loads.Due to complex factors such as climate,economy,and holidays,there are many random uncertainties in power loads based on periodic patterns.Therefore,the study of short-term load in distribution networks based on typical daily load curves is of great significance.With the widespread application of artificial intelligence,information intelligence technology has provided new means for high-precision power load forecasting research.This article selects the typical daily load curve of the power system every month as the indicator for power load forecasting,adopts an improved k-means clustering algorithm,and conducts research on the selection of typical daily load curve,determination of the optimal clustering number K,and prediction algorithm along the data preprocessing,clustering analysis,and prediction analysis paths.The main work is as follows:(1)A study was conducted on the selection method of conventional typical daily load curves.This article uses Spearman correlation coefficient to select the actual day that is closest to the mean vector as this type of typical daily load curve,and selects the typical day with the highest correlation with the monthly benchmark daily load curve for analysis,as a typical daily load curve for that month.(2)Verify the feasibility and accuracy of the deep difference method by comparing several traditional K-value selection methods.(3)In order to compensate for the shortcomings of the k-means algorithm,this article studies the DBSCAN and AHC clustering algorithms,and aggregates and optimizes the k-means clustering with DBSCAN and AHC,respectively.The clustering results are compared with examples to verify the advantages of the k-means AHC algorithm.(4)We conducted research on gas sensitive and non gas sensitive power loads respectively,and provided prediction plans based on typical daily clustering results.