Electricity consumption behaviour analysis based on adaptive weighted-feature K-means-AP clustering

The promotion of smart meters has led to the exponential growth of monitoring data. Consequently, challenges have emerged for utilities to analyse and interpret big data at the demand side. In this study, an adaptive weighted-feature K-means-affinity propagation (AP) clustering algorithm is proposed to analyse customer electricity consumption behaviours. First, a comprehensive feature set is constructed, i.e. the time-domain, frequency-domain and fluctuation features of load curves. Second, the feature set is applied to a divide-and-conquer framework to analyse customer behaviours. In the local modelling, adaptive K-means algorithm is adopted to cluster load curves by considering the time-domain and fluctuation features; the former is objectively weighed by entropy weighting. In the global modelling, AP clustering algorithm is introduced and clustering results are obtained by combining the weighted time-domain and frequency-domain features. The performance tests on the simulation data and electricity consumption behaviour analyses on the big dataset are conducted to verify the effectiveness of the proposed models and approaches. The divide-and-conquer framework is practical, and the proposed features are not only beneficial for clustering the load curves, but also conducive to consumption behaviour analysis.