Research on Text Classification of Large-Scale Power Outage Accidents Based on Machine Learning

As large-scale power outage accidents grow increasingly complex, their textual descriptions exhibit characteristics of multi-factor coupling and semantic diversity. Traditional manual analysis is limited by low efficiency, significant subjective bias, and difficulty in handling unstructured data. This study aims to construct a text classification system for large-scale power outage accidents based on the Cyber-Physical-Social System (CPSS) framework and machine learning, enabling automated identification and analysis of the primary causes of accidents. Based on the CPSS framework, the study categorizes accident causes into three labels: physical layer, information layer, and social layer. It adopts the BERT model for classification, optimizes training and adaptation strategies according to textual features, and compares performance with traditional methods such as decision trees and random forests. Experiments show that the optimized BERT model significantly outperforms traditional methods in accuracy, precision, recall, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score, with the strongest performance in classifying social layer texts. This highlights its robust ability to capture and identify descriptive and management-related semantics. The study contributes to improving the efficiency of accident information processing and provides technical support for power grid emergency decision-making.