Terranova collaborates with Imperial College London to create a new framework for identifying anomalies in distribution networks

The Goal

We set ourselves the goal of identifying frauds, leakages or tampering in gas networks in real time, in order to be able to manage the network efficiently and safely. This is a particularly complex challenge, because it requires the timely processing of large volumes of data from geographically distributed devices, as well as the ability to quickly separate anomalous events from simple operational variations. Timing is in fact a critical factor: the more time elapses between the event and its identification, the greater the potential damage to the network, both in terms of gas leakage and operational inefficiencies, and in terms of safety for people and plants. Detecting an anomaly even a few minutes earlier can make the difference between a targeted intervention and a systemic risk. In addition, the ability to prevent and counteract frauds or tampering makes a decisive contribution to the economic sustainability of the system and user confidence in the service.

The Project

With the students from Imperial College London, we developed a framework, which we called Lunar (Logic Based Utiliy Network Anomaly Recommission). Lunar's main goal is to identify anomalies in distribution networks.

This project has also produced a software tool that is available free of charge, albeit in an absolutely experimental form, using the data provided by the network detected by the sensors to build a model of the network itself. Specifically a model that describes a correct and normal operation of the network - for example in terms of the pressure and temperature values of the gas in the network; where the data is collected from the EE network. Then this tool builds specifications describing the correct operation. At the time of use, when new data is read from the sensors, it is checked against the specifications that were extracted in the learning phase, and if there is any abnormal value compared to the specifications we have, it is detected.

An anomaly is identified by the tool and is proposed to the technician who is picking up the system, who then has the tools to determine whether it is a real anomaly or simply values that have not yet been observed. During the monitoring phase, the tool can also be updated as new data is collected from the network. And if a behaviour is considered normal, the specification is updated to take this behaviour into account as well.