A non-intrusive approach for classifying residential water events using coincident electricity data

This study evaluated the potential for circuit-level electricity data to improve performance by a water end-use disaggregation tool. Support vector machine classifiers were employed to categorize observed water events from an extensive dataset published in the literature. Additional electricity-related event features were assigned depending on temporal proximity to recent clothes washer or dishwasher events. Classifiers were trained on a portion of the dataset with and without the electricity-related features, then tested on an equally sized portion of the dataset. A classifier also categorized events from the testing dataset where event durations were adjusted to match larger sampling intervals, from 10s up to 120s. Specific electricity-related features significantly improved classifier performance for clothes washer, dishwasher, and shower events. Classifier performance was maintained for longer events as sampling frequency decreased, although performance for short duration events decreased. Overall, these results indicate significant potential benefits from integrating electricity-related features for water disaggregation tools. •Support vector machine models were trained to classify residential water events.•Features related to coincident electricity use by mechanical devices were defined.•The tradeoff between data sampling interval and classifier performance was explored.•Performance was judged via confusion matrices and receiver operator characteristics.•Results indicated electricity-related features assist water event disaggregation.