Predicting Thunderstorm-Induced Power Outages to Support Utility Restoration

摘要: Strong thunderstorms have substantial impacts on power systems, posing risks and inconveniences due to outages. Developing models predicting the outages before a storm is high priority support restoration planning. However, most outage data are zero-inflated, which results in some challenges predictive modeling such as bias inaccuracy. Power also stochastic there always exists irreducible variability predictions. The goal develop overcome caused by zero-inflation accurately estimate terms of probability distributions better address inherent stochasticity uncertainty This paper proposes novel approach integrating mixture with resampling cost-sensitive learning for distribution number Validating using data, we demonstrate that our offers more accurate point probabilistic predictions compared traditional approaches, supporting utility