Multiscale fractal-LLM synergy for secure synchrophasor data authentication in smart grids

Synchrophasor Data (SD), which captures the states and dynamics of transmission and distribution power networks, is recorded by dispersed Phasor Measurement Units (PMUs) that are connected through advanced communication infrastructures in cyber-physical systems. While the source information of SD is critical for PMU-based applications, it remains susceptible to data integrity attacks. Reliable SD source authentication must be performed efficiently before any SD analytics to ensure application integrity. Existing machine learning authentication methods are focused on sparse SD across vast regions and fail to represent realistic network conditions. This study presents a data-driven SD source authentication of dense PMUs at intra-state locations without requiring detailed power network topology, parameters, or costly upgrades to PMU hardware and data infrastructure. It offers a practical data-driven approach through multiscale fractal - Large Language Models (LLMs) synergy for secure SD authentication. Specifically, Multiscale Vector Visibility Height Correlation Analysis (MVVHCA) is developed to extract long-term fractal correlations from SD with multiple attributes simultaneously and pinpoint scales exhibiting peak multifractality. Then, high-fidelity synthetic SD is reconstructed through Adaptive Multifractal Interpolation (AMFI). These synthetic SD are further processed by LLMs to achieve robust, real-time source authentication. Validation on extensive real-world SD from multiple intra-state locations confirms superior accuracy and computational efficiency, which establishes a scalable solution for secure operation of renewable-dominated smart grids.