Persistent Homology-Based Clustering to Extract Representative Load Profile From Aggregated Residential Customers

Data collected by smart meters are frequently used to identify representative profiles for various applications. However, changes in the electricity grid infrastructure and market alter the consumption behavior of residential customers, making it more complex and random, and therefore more difficult to extract accurate profiles. This article proposes a persistent homology-based method to extract the representative load profile from a collection of residential aggregated load datasets. Homology-0 is used to build a graph with the most persistent data structure. Subsequently, a modularity-based algorithm is employed to extract the representative profile and remove atypical days. Comprehensive comparisons with specialized and neural network-based methods are performed to test its effectiveness. The results are evaluated with a three-index-based quantitative analysis employing Silhouette, Davies–Bouldin, and Calinski–Harabasz indices, and some qualitative visual tools are also included. The nonparametric dependence and assumption-free nature of the proposed method make it practical for the new challenges facing the electrical distribution network.