Integrating Structural Vulnerability Analysis and Data-Driven Machine Learning to Evaluate Storm Impacts on the Power Grid

Abstract

The complex interactions between the weather, the environment, and electrical infrastructure that result in power outages are not fully understood, but because of the threat of climate change, the need for models that describe how these factors produce power grid failures is acute. Without them, it remains difficult to understand the amount of weather-related damage we may expect in the future, as well as how changes or upgrades to the infrastructure may mitigate it. To address this problem, a modeling framework is proposed in this article that integrates data derived from structural vulnerability analysis into a machine-learning based weather-related power outage prediction model to create a model that is sensitive both to the weather and the technical configuration of the infrastructure. This Physics Informed Machine Learning (PIML) approach is demonstrated using data from a major power utility operating in the US State of Connecticut, and is compared against a fragility curve modeling approach using some of the same data. The validation of the PIML model shows superior predictive ability, as well as variable sensitivities that follow expected patterns. These results suggest that the model would be able to evaluate the influence that different configurations of the infrastructure would have on the occurrence of power outages caused by severe storms, allowing for the anticipated effects of investments in infrastructural upgrades to be quantified and optimized.

Keywords

• Electrical distribution
• fragility curves
• machine learning
• power grid
• power outages
• reliability