Prediction and mitigation of nonlocal cascading failures using graph neural networks

Area of Science:

• Complex systems analysis
• Electrical engineering
• Network science

Background:

• Cascading failures in power grids propagate nonlocally, posing significant risks.
• Existing mitigation strategies require careful formulation due to unpredictable failure propagation.

Purpose of the Study:

• To propose a novel strategy for mitigating cascading failures in electrical power grids.
• To introduce Avalanche Centrality (AC) as a measure for node impact on avalanche dynamics.
• To develop an efficient computational method for AC calculation in large networks.

Main Methods:

• Introduced Avalanche Centrality (AC) to quantify node influence on cascading failures.
• Employed graph neural networks (GNNs) for efficient AC prediction in large-scale networks.
• Validated the strategy by identifying and reinforcing high-AC nodes for mitigation.

Main Results:

• Avalanche Centrality (AC) effectively reduces avalanche size compared to heuristic measures.
• Graph neural networks (GNNs) enable computationally efficient AC prediction for large power grids.
• The proposed strategy allows for effective mitigation of cascading failures in complex networks.

Conclusions:

• The developed framework offers an efficient method for mitigating cascading failures in electrical grids.
• The approach using Avalanche Centrality and GNNs is adaptable to other complex systems with simulation challenges.