Overview

The DIADEM project aims to develop a framework for the sustainable use of energy and drinking water by supporting smart distribution systems with data driven managing tools. The main objective is to define and build up an innovative data driven methodology to support the distribution operators in managing smart networks to guarantee resilient systems, efficient services and suitable use of water and energy resources. Specifically, the project focuses on the anomaly detection of data provided by smart meters for identifying early problems in Water Supply Systems (WSS) and District Heating Systems (DHS) such as pipe bursts, regulating devices failure and unaccounted consumption. The project comprises different activities starting from the issue of raw data collection to the validation of machine learning models for anomaly detection.

Data driven methods crucially rely on large amounts of data. Therefore, information from smart meters installed in grids will be collected. Furthermore, a generator of synthetic data will be developed, in order to increase the variety of data to train machine learning models. A pre-processing procedure will be proposed that includes data filtering, gaps imputation, outliers identification, temporal aggregation definition and preliminary analysis, e.g. performing analysis of serial correlation and correlation with exogenous variables.

Different types of data driven models among statistical and machine learning methods will be tested and analyzed, and then an ad hoc method will be proposed, based on the best performing techniques to deal with multivariate time series with uncertain data. In particular, a prediction model will be coupled with a classification model, to classify different types of anomalies based on the deviation between predicted and measured data. An important goal of the proposed machine learning method is to minimize false alarms. Hence, the classification model with be specifically tailored to this end.

The developed framework will be tested on Egna and/or Laives WSS and Renon DHS for validating the proposed approach in supporting the local operators to enhance the reliability of their service and to avoid water and energy waste.

This project is a collaboration between the Faculty of Science and Technology and Faculty of Computer Science of the Free University of Bozen-Bolzano.