Time-Series Adjusted FasterPAM Clustering of Enterprise Electricity Consumption Based on Autocorrelation Function

Analyzing enterprise electricity consumption behav-ior and patterns is crucial for studying the power market. One approach is to use time series clustering to analyze enterprise electricity consumption, allowing for the customization of services to enhance satisfaction for enterprises with similar characteristics and aiding in the detection of abnormal usage to improve electrical safety. However, the significant differences in scale among different enterprises' electricity consumption make direct clustering challenging. Traditionally, features of time series, such as autocorrelation functions, are extracted to classify time series with similar structures. Selecting different lag numbers for the autocorrelation function can lead to varying results, making this selection critically important. This paper proposes a novel time series clustering method based on estimated autocorrelation functions, which automatically selects the appropriate lag number and integrates with the FasterPAM clustering method to enhance computational efficiency. Moreover, since the FasterPAM method requires predefining the number of clusters, we introduce a new index to determine the optimal number of clusters. Thus, the proposed autocorrelation-based adjusted FasterPAM clustering method offers a more automated approach. Finally, we apply this method to real-world data, clustering the monthly electricity consumption of 197 enterprises in a certain province of eastern China, and analyze the structural characteristics of the clustered time series.