Online Water Usage Monitoring under Anomalous Interference in Residential Households

As the issue of water shortage is increasing nowadays due to climate change, water consumption monitoring has become more critical in home automation services in recent years. In order to lower water bills, residents need to adjust their water usage behaviors to reduce their water consumption, highlighting the importance of the water behavior disaggregation task. However, existing works may fail to precisely disaggregate behaviors when anomaly data exists in received water data since they usually assume it is a clean dataset. In order to deal with this issue, we propose a two-phase framework to online disaggregate water usage behaviors in consideration of the occurrence of water anomaly data. A density-based clustering and different pretrained classification models are combined to detect anomalies efficiently and effectively recognize different usage behaviors. As studied on the real-world dataset, we demonstrate that the proposed framework can achieve good performance on datasets with or without anomalies.