A Mixed Strategy Based on Self-Organizing Map for Water Demand Pattern Profiling of Large-Size Smart Water Grid Data

Abstract

In the present paper a procedure is introduced to detect water consumption patterns within water distribution systems. The analysis is based on hourly consumption data referred to single-household flow meters, connected to the Smart Water Network of Soccavo (Naples, Italy). The procedure is structured in two consecutive phases, namely clustering and classification. Clustering is performed on a selection of standardized monthly time series, randomly chosen within the database; different clustering models are tested, basing on K-means, dendrogram and Self-Organizing Map, and the most performant is identified comparing a selection of Clustering Validity Indices. Supervised classification is performed on the remaining time series to associate unlabeled patterns to the previously defined clusters. Final results show that the proposed procedure is able to detect annual patterns describing significant customers behaviors, along with patterns related to instrumental errors and to abnormal consumptions.