Incremental Electricity Consumer Behavior Learning Using Smart Meter Data

Electricity consumption behavior features can be represented by the load patterns extracted from daily load data clustering. Such representative load patterns should be updated because consumer behaviors may be changed over time. We propose a novel incremental clustering algorithm with probability strategy, known as ICluster-PS, to update load patterns without overall daily load curve clustering. Given the existed load patterns and new daily load data, the method first extracts new load patterns from new data, and then intergrades the existed load patterns with the new ones. Finally, an addition modification is performed on the intergraded sets to obtain the optimal updated load patterns. Several essential parameters are updated after this procedure so that the algorithm can be performed continuously. Extensive experiments are implemented on real-world dataset. The results are evaluated by both accuracy measures and clustering validity indices, which indicate that our method can provide an efficient response for electricity consumption patterns analysis to end consumers via smart meters or other facilities with resource constraints.