Time series grouping algorithm for load pattern recognition

Abstract

System analysis and real-time operations in power distribution utilities require an accurate but compact load data model created on the basis of large number of consumers’ measurements modeled as high-dimensional time series. This paper proposes an algorithm for grouping time series with similar load patterns and extracting characteristic representatives of loads from the obtained groups, resulting in reduced load data model size. The proposed Time Series Grouping Algorithm combines dimensionality reduction, both partitional and hierarchical clustering and cluster validation to group time series into an optimal number of clusters based on simple parametric settings. The usefulness of the proposed algorithm is proven in a case study implemented in R language. The case study was conducted on real smart meter data from three distribution networks: one North American and the other two European. Results of the case study confirm that the proposed solution achieves high cluster validity and short execution time comparing to related algorithms. Therefore, the article’s main contribution is load pattern recognition support convenient for applications in distribution management systems.