Active learning and relevance vector machine in efficient estimate of basin stability for large-scale dynamic...

Area of Science:

• Electrical Engineering
• Network Science
• Computational Science

Background:

• Dynamic networks, like power grids, face cascading failures due to interdependencies.
• Basin stability (BS) quantifies the reliability of these systems, crucial for grid resilience.
• Identifying vulnerable nodes is key for optimal maintenance and preventing blackouts.

Purpose of the Study:

• To develop a computationally efficient method for estimating basin stability (BS) in large-scale power grid networks.
• To reduce the reliance on computationally expensive Monte Carlo (MC) simulations for N-1 reliability assessments.
• To improve decision-making for power grid maintenance by accurately identifying vulnerable components.

Main Methods:

• Utilized a classification approach reframing BS estimation.
• Investigated the application of relevance vector machines and active learning.
• Focused on efficiently locating the boundary of stable dynamics (basin of attraction).

Main Results:

• Achieved over 95% reduction in simulation cost for N-1 reliability assessment.
• Demonstrated an efficient alternative to traditional Monte Carlo methods.
• Successfully applied machine learning to complex network reliability problems.

Conclusions:

• The proposed machine learning approach offers a significant computational advantage for power grid reliability analysis.
• This method enables faster and more cost-effective identification of critical vulnerabilities in power systems.
• Enhanced BS estimation facilitates proactive maintenance strategies, improving overall grid stability and security.