Metering Asset Fault Perception and State Assessment Model Fusing Multi-modal Visual Learning

The intelligent transformation of power grids imposes higher requirements for real-time state monitoring and precise fault diagnosis of metering assets. However, existing big data-driven methods face significant challenges that severely weaken model generalization capability and assessment accuracy. Fusing multimodal visual learning algorithms can break through these bottlenecks, providing support for high-precision, highly reliable fault perception and state assessment. To this end, this paper proposes a metering asset fault perception and state assessment model fusing multimodal visual learning. Its core is an innovative dynamic adaptive fusion algorithm, which effectively solves the problem of response lag in traditional static fusion through real-time weight adjustment and interference filtering mechanisms, significantly enhancing system robustness. The model built based on this algorithm covers the entire processing chain: multi-modal data input, feature decoupling, adaptive fusion, fault classification, and state scoring. Finally, the accuracy and practicality of the proposed model are verified using an actual metering device fault sample library from a provincial grid.