Clustering electricity consumption patterns using functional data analysis

Investigating energy usage patterns in the commercial sector and identifying notable characteristics is imperative for implementing more flexible and effective demand-side management strategies, reducing energy costs, and improving energy efficiency. Properly clustering energy consumption patterns enables us to identify and distinguish major consumer groups and their energy load characteristics. This paper uses aggregated 3 years of smart meter data from approximately two thousand commercial customers to find major load profile clusters. Functional data analysis (FDA) is applied to the monthly aggregated power usage data to capture the dynamic and functional nature. The results show that the general amount of electricity usage, corresponding to the first functional principal component (FPC), dominates the function-to-function variability, and most of the remaining variability can be explained by three additional curve shape features, corresponding to the second through fourth FPCs. To account for the largely different scales and nonhomogeneous densities of the clustering variables, which are FPC scores, a multi-level nested clustering, a combination of the Gaussian mixture model and clustering tree, is performed. The resulting clusters are summarized by FPC scores, which easily characterize their consumption patterns, demonstrating the primary advantage of FDA.