In this study, we address the evolving cybersecurity threats facing modern power grids. Existing approaches often rely on discrete label–based classification and single-modality analysis, which fail to capture the continuous nature of threat intensity and the heterogeneity of power grid data. To overcome these limitations, we propose PGTSAGA (Power Grid Transformer for Security Situational Awareness and Adaptive Grayscale Assessment), a novel Transformer-based framework for multimodal power grid security analysis. PGTSAGA integrates SCADA measurements, PMU synchrophasor data, and network traffic data through a Trimodal Cross-Attention Mechanism, enabling effective multimodal feature fusion. A Hierarchical Transformer Architecture extracts threat features across multiple temporal and relational scales, from local anomalies to global grid conditions. Furthermore, a Continuous Threat Evaluator based on variational inference models threat intensity as a probability distribution, capturing uncertainty in noisy data. Complementing this, an Adaptive Grayscale Classification method grounded in fuzzy set theory dynamically maps threat levels into a continuous grayscale space, reducing errors caused by hard discrete classification. Experiments on real-world datasets demonstrate that, compared with Crossformer, PGTSAGA achieves a 5.4 % relative improvement in Accuracy, approximately 6 % increases in Precision, Recall, and F1, a 4.12 % increase in AUC, and a 40.74 % relative reduction in the false alarm rate.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号