Predicting Imminent Electrical Safety Incidents Using Smart Meter Big Data With Large Language Models

Abstract

This study examines the application of Large Language Models (LLMs) in predicting imminent electrical safety incidents using extensive smart meter data. The proliferation of smart meters in residential settings has led to the generation of large-scale datasets, which provide valuable insights into household electricity consumption trends. However, early detection of hazardous electrical conditions, such as overloads or short circuits, remains an arduous challenge. This research proposes a novel methodology that utilizes LLMs to analyze real-time electricity usage data, employing a defined text-based encoding of numerical values to identify potential electrical hazards. By training the model on historical consumption data from over 700 households, this approach aims to uncover subtle patterns that may indicate precursors to unsafe electrical conditions. Results indicate that LLMs significantly improve predictive accuracy for identifying such incidents, thus offering a proactive strategy for mitigating electrical safety risks. This work contributes to the fields of energy management and electrical safety by equipping utility providers and homeowners with early warnings and actionable insights that support the prevention of electrical safety accidents.