Research project

Industrial Artificial Intelligence for Safety in Gas Networks

Self-learning tools for analysis, prognosis and decision support

Project goal

Strengthening the resilience of gas networks

The focus is on resilient network management in critical situations such as natural disasters, geo-political tensions or attacks on a grid. In particular, reinforcement and adversarial learning will be used for anomaly detection and operator support.

Operators are supported by means of continuous state evaluation and anomaly detection in order to reliably identify the need for early action. Actions for optimized operating modes can thus be identified.

Operators can react appropriately to unusual network states, enabling efficient and sustainable operation. Cascading effects must be avoided.

New requirements for gas grids

Integration of renewable energies

Gas network operators must ensure efficient transport options both for large volume hydrogen imports and for decentralized and volatile inlets from their own biogas and hydrogen generation plants.

In the future, gases in various compositions with different calorific values must be transported safely and efficiently. The basis for this is precise modeling of the chemical and thermodynamic properties of the gases.

A current challenge is the transformation of gas grids for the integration of renewable energies into the overall energy system. The focus here is on the development of pure hydrogen networks as well as the conversion of the existing gas infrastructure.

Software with advanced features

Reinforcement and adversarial learning

PSI supports network operators adapt to the changing market conditions. They integrate their highly specialized knowledge of network operators in its software solutions in order to contribute to a safe, environmentally compatible and efficient energy supply.

Based on our robust software PSIgasguide and PSIganesi, the use of machine learning is being further tested. Reinforcement learning is based on PSI's own Deep Qualicision technology.

This is intended to ensure optimal operation of the gas grids, even with volatile inlets of gases with different properties.

Supply reliability

Recognize critical situations and manage them effectively

Fast and safe access to experienced know-how is crucial. The mastery of all control actions from normal operation as well as the training of critical situations is required.

The specialized knowledge of experienced operators for the evaluation of exceptionally critical situations and the control actions based on this knowledge is facilitated with the help of CI systems.

Network dynamics are learned by a machine learning algorithm, depending on the grid state and control actions. Anomalies in regular operation can thus be detected reliably and at an early stage.

Project Partners