Utilization of Machine Learning to Detect Sudden Water Leakage for Smart Water Meter

Abstract

This article deals with the use of machine learning to detect sudden water leakage. A smart water meter, which enables monitoring the water consumption of the observed object, is used as the source of input data. Based on these data and their analysis, a symbolic regression, which must know not only the input parameters but also the structure of the model, was finally used to build the model. After finding a suitable function and standard deviation from the model, it is possible to set the required sensitivity and thereby detect anomalous states of water consumption in monitored time windows. Since the smart water meter also has a ball valve, if a sudden water leakage is detected, the water meter can autonomously close the main supply and thus avoid extensive damage.