Modeling of water distribution networks and AI algorithms for anomaly detection (with UNIACQUE)

Leakage in the water supply networks is becoming one of the largest resource waste of the system, ranging from the 10-20% of the total water for the newly updated infrastructure up to nearly 60% in remote places like mountain villages.

Since water is a fundamental and limited resource in the society, the supplier of the province of Bergamo, Uniacque s.p.a, has started a project, in collaboration with the University of Bergamo, that aims to update their supply network using more advanced telemetry sensors that allow to monitor in real-time the state of the system.

This industry 4.0 project allows to collect a huge amount of water consumption data that can be used for various analysis. The previously cited leakage detection is a good example of what it’s possible to do with these data. Other examples are fraud detection or user classification.

The CAL is involved in the analysis of this huge amount of data. A preliminary software for the complete monitoring of the network has been developed and some preliminary real data were analyzed.

Currently, Uniacque is starting a pilot project in a small city in the province of Bergamo

Figure 1: Example of the developed system that monitors the state of the network. This simulation was run using real water consumption data associated with each simulated user (the dots at the extremes). The simulation allows to estimate the pressure value at each node, the flow in each tube, the water level in the tanks and many other hydraulic measures of the network.
Figure 2: Preliminary results of a clustering algorithm on real water consumption data. The algorithm automatically divides the users between three distinct categories: Inactive users, sporadic users and active users. On the right, there is reported an example from each cluster.