Unveiling Weather-Induced Blackouts: A Ten-Year Analysis With Deep Learning-Driven Power Resilience Enhancement

Abstract

When a rainy day affects the power grid, instead of enjoying the weather, many consumers face unplanned blackouts worldwide. Approximately 80% of blackouts in the US are weather-induced power outages. Through the amalgamation and meticulous preprocessing of diverse public datasets, encompassing variables such as maximum temperature, solar exposure, and precipitation levels, the study aims to unravel the intricate dynamics through which weather influences power resilience. We utilize over ten years of data from 47 local government areas. The analysis focuses on predicting future power outages using a state-of-the-art deep learning Long Short-Term Memory (LSTM) model. The results show a promising area under the Receiver Operating Characteristic (AUC ROC) curve of approximately 90% and a mean precision exceeding 96%. The experiments utilize a 5-fold cross-validation methodology to ensure robustness and reliability in the predictive model. It reveals the nexus between weather patterns and power systems and offers practical insights. The proposed work can serve as a valuable resource for all stakeholders in the energy sector, fostering informed decision-making and contributing to the ongoing dialogue on enhancing power resilience, improving cyber-physical infrastructure, and disaster preparedness.

Keywords

• Weather-induced power outages
• power grid resilience
• power outage prediction
• deep learning
• long short-term memory