The evolution towards modern power systems necessitates enhanced security protocols for human-machine interactions, particularly in scenarios like mobile patrolling and dispatching operations, where conventional authentication methods are vulnerable to forgery and lack reliable traceability. This paper addresses this critical challenge by introducing Dual-AngleNet, an interpretable and lightweight CNN-LSTM model that authenticates personnel identity by leveraging the unique and difficult-to-replicate neuromuscular characteristics of pen-tip tilt dynamics. The model integrates azimuth and altitude angles for robust spatiotemporal feature learning. Its lightweight architecture is specifically designed for practical deployment on edge devices within the power grid ecosystem. Furthermore, the combination of a hierarchical attention mechanism and SHAP analysis ensures that the authentication decisions are transparent and interpretable to system auditors, fostering trust. Experimental results on a dataset of 600 signature samples confirm the model’s superior accuracy (97.69%) and resilience (2.18% EER) against forgery compared to existing methods, while maintaining efficient inference suitable for edge deployment. This research proposes a methodological foundation for building a trustworthy, transparent, and operationally-integrated identity authentication framework applicable to critical infrastructures like the future power grid.
Hao et al. (Mon,) studied this question.