A combination of multi and univariate anomaly detection in urban irrigation systems

Water scarcity is a global concern that requires a more intelligent way to manage the water network in order to minimize consumption and losses. There is a current need for new automated and intelligent decision support systems that employ data analytic advances to analyze in real-time the anomalies and problems that appear in water usage. In our case, we have developed a 3 steps system using unsupervised algorithms, that has been tested on the irrigation of urban parks. Those steps consist of grouping the parks according to similarities in the irrigation by clustering the consumption time series, searching for the dates where multivariate anomalies occur in every group using the Vector Autoregressive Model, and finally, applying an ARIMA framework to each series in the area of those dates. Our methodology reduces the time that anomaly detection univariate systems require for analysing the whole univariate time series by extracting previous knowledge on a multivariate approach. This methodology, when integrated with an IoT platform, is a tool for easing the labelling of real anomalies and can help create supervised datasets for future research in the area. The approach is used in urban scenarios, however, can easily be extended to be an application for smart agriculture scenarios.