Computationally efficient tail distribution-aware large-scale power system overloading risk assessment.

Area of Science:

• Power Systems Engineering
• Computational Electromagnetics
• Risk Analysis

Background:

• Renewable energy sources (solar, wind) introduce intermittency and uncertainty.
• This variability poses significant overloading risks to power systems, potentially causing cascading failures.
• Quantifying these risks in large-scale power systems is computationally challenging.

Purpose of the Study:

• To develop a computationally efficient and accurate method for quantifying overloading risks in large-scale power systems.
• To address the challenges posed by intermittent renewable generation and N-k contingencies.
• To improve the fidelity of risk assessment near the overloading threshold.

Main Methods:

• A deep-kernel sparse vector-valued Gaussian process (GP) was developed as a surrogate model.
• The GP model incorporates generation dispatch, contingencies, and uncertain inputs (photovoltaic power, load demand).
• An adaptive resampling mechanism using a power flow solver was introduced to correct surrogate model biases.

Main Results:

• The proposed method accelerates risk assessment by 22 times compared to Monte Carlo sampling on a 21k+ bus system.
• High accuracy was maintained in overloading risk quantification.
• The approach demonstrated robustness across various distribution types and correlation scenarios.

Conclusions:

• The developed surrogate modeling approach provides a computationally efficient and accurate solution for power system risk assessment.
• The adaptive resampling mechanism enhances the fidelity of predictions near the overloading threshold.
• The method is effective for large-scale power systems with significant renewable integration.