Research On Key Technologies For Load Pattern Shape Clustering

The Manufacturing industry is dominant sector in electric power consumption.Currently,advanced Metering infrastructure is widely deployed to collect real-time customer electricity consumption data.The knowledge of customer electricity consumption profiling can be used to smart grid dispatching,marketing and pricing based on analyzing and mining the load profiling in different industries,seasons and time periods.This dissertation proposes two algorithms to analyze load pattern shape(LPS)of manufacturing users,which can discover typical LPS modes and power consumption patterns.Clustering is an essential methodology for LPS mining.However,there are some critical challenges for the existing LPS clustering algorithms,such as,how to select the optimal number of clusters in LPS results? How to overcome the performance bottleneck of the classical single clustering algorithm? How to find out the optimal segmentation of feature spaces for high dimensional load pattern shape(LPS)data?In this dissertation,the solution for LPS data clustering explored in the following issues.First,we propose an extended fuzzy K-Means algorithm,which automatically discovers the optimal number of clusters,can find out the typical power consumption patterns in different industry sectors.Secondly,we design an ensemble clustering algorithm with stratified feature sampling,rather than the classical single clustering algorithm.The proposed algorithm generates lots of subset candidates using stratified features sampling scheme.The candidates clustering results can merge to the final optimal result with MCLA hypergraph conjunction function.Finally,we apply an innovative histogram cut strategy based on features differential accumulation to identify the feature span in LPS data.This approach can significantly enhance the performance of the proposed ensemble clustering algorithm for high dimensional feature division.There are different experiments for the historical power consumption data from more than 20,000 manufacturing factories in Dong Guan in South China PowerGrid.The results prove fuzzy K-Means extension algorithm can find out the typical LPS mode in different industry sectors,such as two shift-term,three shift-term or night shift-term,etc.The features stratified sampling ensemble clustering algorithm overperformed the classical clustering algorithm,even in a low sampling ratio and noisy data.All these results show the proposed algorithms can discovery the typical LPS mode and power consumption patterns efficiently,which can provide a valuable solution for demand-side management and dynamic peak electricity price calculation.