Electricity Consumption Patterns Research Based On Machine Learning

In the last decade,many developing countries have devoted themselves to promoting the construction and development of smart grids.Compared with traditional power grids,the application of smart grids can determine and classify users’ electricity consumption patterns through user electricity consumption behavior identification algorithms,which is of great significance for realizing demand response and electricity price setting.Aiming at the problem of feature selection strategy of electricity consumption data in K-means algorithm,this paper proposes an iterative feature selection strategy that considers the correlation between features.First,based on the K-means clustering effect of different feature combinations based on the DBI index as the feature selection criteria;then,establish a feature optimization model,select the candidate features through experiments,and then add the preferred feature combination;finally,based on specific types of users.The feature optimization of electricity consumption data was realized,and the optimization analysis of K-means clustering algorithm was completed.The actual power consumption data of the power grid verifies the effectiveness of the proposed strategy to improve the clustering accuracy and reduce the computational complexity.It’s very important for electric energy related organization to categorize the electricity consumption behavior of consumer,in order to realize the consumer’s requirement and accomplished them with satisfy.For classification of consumption patterns k-means clustering is one of the best method.Although,the initial clustering centroids in the traditional K-means algorithm are randomly chosen,which creates it susceptive to local optima and has complexity in concentric to the global minimum.On the basis of the above feature optimization,the K-means algorithm randomly selects the initial clustering center and artificially presets the number of clusters to cause poor stability of the algorithm,and proposes the initial clustering center selection and the splitting strategy between clusters.Experiments show that the algorithm has better stability than the traditional K-means algorithm,and can be iterated on the basis of the existing clustering results,saving computing resources.