IKIGas

Industrielle Künstliche Intelligenz für Sicherheit in Gasnetzen (Industrial artificial intelligence for safety/security in gas networks). In particular, adversarial reinforcement learning to uncover safety gaps and unsupervised anomaly detection, for a safe gas network.

Funding Agency

Funding for the topic area "Artificial Intelligence in Civil Security Research II" as part of the Federal Government's "Research for Civil Security 2018 - 2023" program.

https://www.sifo.de/sifo/de/projekte/querschnittsthemen-und-aktivitaeten/kuenstliche-intelligenz-in-der-zivilen-sicherheitsforschung/kuenstliche-intelligenz-in-der-zivilen-sicherheitsforschung_node.html

Goal/description

The overall IKIGas project aims to pave the way for gas network operators to use artificial intelligence to quickly identify threats and dangerous situations in the gas network and remain capable of acting in a comprehensible manner. The individual goal of the THI sub-project is to combine existing real and simulation data from gas network operations in order to train adversarial and benign AI agents with the help of reinforcement learning and to recognize a wide range of anomalies using unsupervised learning methods. These software agents deliberately generate data for critical situations by manipulating the environment and the gas network. For unsupervised anomaly detection models, test cases based on time series data (e.g. pressures, linepacks, control commands) will also be created to differentiate between valid and invalid processes or states.