Meteorological Factor-Based Early Warning of Power Grid Faults via the Meta-Fuse Framework

INTRODUCTION: Extreme weather events increasingly threaten power grid resilience, causing cascading failures, while conventional models fail to capture complex interactions among meteorological, topological, and spatiotemporal factors. OBJECTIVES: This paper aims to develop Meta-Fuse, a meteorological attention-based fused model, for early warning of power grid faults under environmental uncertainty. METHODS: This study utilizes methods grounded in the analysis of social networks. The proposed framework integrates Triangular Topology Aggregation Optimization, a CNN-BiLSTM pipeline for spatiotemporal feature extraction, and a multi-head attention classifier, fused via a hybrid CRITIC-entropy strategy. RESULTS: The model achieves a classification accuracy of 0.9464 with incomplete weather data and maintains a sensitivity of 0.9309 with high-dimensional features, outperforming traditional architectures. CONCLUSION: Meta-Fuse demonstrates effective topology-informed learning and multimodal fusion, confirming its value for proactive resilience strategies in weather-exposed power systems.