Hybrid learning-based fault prediction and cascading failure mitigation in multi-network energy systems

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence
• Network Science

Background:

• Increasing integration of electric, gas, and heating networks elevates the risk of cascading failures.
• Robust models are essential for predicting and mitigating fault propagation in complex, interconnected energy infrastructures.

Purpose of the Study:

• To develop a novel AI-based management architecture for predicting and mitigating fault propagation in multi-energy networks.
• To enhance the resilience and reliability of interconnected energy systems through adaptive recovery strategies.

Main Methods:

• Coupling adversarial learning with graph-structured predictive models for fault evolution synthesis and spatiotemporal correlation capture.
• Utilizing a generative network for fault pattern synthesis and a graph-based neural network for subsystem correlations.
• Incorporating distributionally robust optimization for adaptive recovery strategy development under uncertainty.

Main Results:

• The proposed AI model accurately predicts fault propagation across interconnected electric, gas, and heating systems.
• Optimized recovery strategies significantly reduce restoration time and minimize the impact of cascading failures.
• Demonstrated effectiveness using a synthetic case study involving IEEE 123-bus electric, Belgian gas, and a heating network.

Conclusions:

• The integrated framework offers a scalable solution for enhancing resilience in modern interconnected energy networks.
• Pioneering use of adversarial and graph-based learning advances fault trajectory prediction in multi-energy systems.
• Distributionally robust optimization strengthens recovery planning, improving overall system reliability.