A Data-Driven BIRCH Clustering Method for Extracting Typical Load Profiles for Big Data

In typical load shape analysis, many different clustering methods have been used to segment customers, interpret behavior and inform marketing reach out strategies. Due to memory requirements and computational efficiency, many clustering algorithms do not have the capabilities to perform analysis at the urban-scale. In this paper, a scalable data-driven BIRCH clustering algorithm is used to extract the typical load shapes of a neighborhood. The BIRCH radius threshold is determined by solving an optimization problem. For global clustering, a metric is created that can rank the best possible options for the agglomerative phase of the BIRCH algorithm. The developed method allows large time series data at the urban-scale to be quickly analyzed.