Smart Water Meter Network Anomaly and Failure Detection System

In the article we present a solution for analyzing water consumption from a multi-family buildings. In the first stage, data from the tested water meters are collected, aggregated and prepared for initial processing. In the next step, we check whether there are any gaps in the time series and, if so, we fill them using interpolation. After this time, the time series is processed by machine learning algorithms in real time. Traffic features from the smart meter network are collected within one minute and processed in sliding time windows. Anomaly or failure conditions take into account the number of times the values of the current signal extracted on the line exceed the prediction interval per unit time. Finally, a report is submitted to the maintenance services about possible problems with a selected part of the water infrastructure. The obtained experimental results confirmed the effectiveness of the presented method.