Load Demand User Profiling in Smart Grids with Distributed Solar Generation

Clustering of customer consumption (load) patterns, known as load demand user profiling, is a technique widely used by utility companies. Load demand user profiling facilitates tasks critical to grid management such as load prediction, design and implementation of incentive programs such as Demand Response (DR), etc. Traditional load clustering algorithms perform poorly in smart grids with significant penetration of distributed Behind-The-Meter (BTM) solar. This is because these clustering algorithms cannot find load only patterns in the net-load data - load minus solar data obtained from the AMI meters of customers with BTM solar installed. These algorithms fail to find similarities between the consumption patterns of the customers in such grids due to the increased unobservability introduced by BTM solar leading to poor clustering quality. However, accurate load clustering is important even in grids with BTM solar for customer behavior modeling which is required for customer incentive programs and load predictions. To address this problem, we develop two novel load clustering techniques that use Consumer Mixture Model (CMM). CMM models the load patterns of customers with BTM PVs as a linear combination of the load patterns of customers without PV. The first technique uses the weights of the linear combination to assign the customers to mean patterns with largest weights while the second technique performs load and solar disaggregation of the net-load data and performs clustering on the disaggregated load data. We compare our techniques with two state-of-the-art clustering algorithms using real life Pecan Street dataset based in Austin, Texas. Our experiments show that our techniques produce clusters of higher quality in terms of the widely used Silhouette Coefficient metric when compared with the baseline algorithms. Baseline algorithms produce clusters with negative Silhouette Coefficient while the clusters produced by our methods have a positive Silhouette Coefficient. Observations of the resulting clusters show that our method is able to group customers of similar consumption patterns together regardless of the presence of BTM solar generation. Index Terms-Time series clustering, Smart grids, Load Profiling, BTM Solar