Assessing grid hardening strategies to improve power system performance during storms using a hybrid mechanistic-machine learning outage prediction model

Abstract

Improvements of power system performance during severe weather events are targeted through grid hardening actions, such as strengthening aging infrastructure or performing vegetation management. However, due to the complexity of modeling power system interactions with the environment and the impacts of proposed reinforcements, the benefits of such hardening actions are often unclear. To support hardening assessments, comprehensive models capturing these intricate relationships are necessary. To this end, this study presents a novel application of a hybrid mechanistic-data-driven outage prediction model (OPM) designed to incorporate changes in infrastructure or environmental/vegetation parameters in the prediction of storm outages. The model incorporates physics-based structural fragility curves of the overhead pole-wire system within a machine learning OPM trained on meteorological, topographic, infrastructure, and historic outage data to quantify the benefits of adaptive change in reducing storm-related power outages. A prioritization scheme is formulated to inform improved strategies to strengthen grid resilience under budgetary constraints. As a case study of the distribution system in Connecticut, the model was trained on historical outage data from storms between 2005 and 2020, and a counterfactual analysis is conducted. The results indicate tree removal and pole strengthening are the most cost-effective strategies; for instance, under a $600 million budget, tree removal could have reduced up to 15,000 distribution grid outages over the events analyzed. The developed framework and models can be useful for utility companies, regulatory agencies, and governments to highlight areas of need, inform cost-risk-benefit analyses, and aid in the optimal allocation of funds and resources toward grid hardening.