Deep learning detection and robust observer for attacks in multi-terminal direct current wind grids

This study focuses on designing Wide-Area Damping Controllers (WADCs) for a Multi-Terminal Direct Current (MTDC) system linked to a wind farm employing Doubly-Fed Induction Generators (DFIGs). These systems are prevalent in modern power grids to manage disturbances. Vital data for WADCs are acquired from geographically dispersed Phasor Measurement Units (PMUs) and channeled through the IEEE C37.118 protocol to Phasor Data Concentrators (PDCs) for time alignment. Malicious Time-Disrupting Synchronization Attacks (TDSAs) can tamper with PMU data, endangering grid stability. The research proposes an advanced TDSA model exploiting vulnerabilities in the PDC’s time-synchronization process to hinder WADC’s monitoring of oscillatory patterns. A strategy for detecting and mitigating TDSAs is introduced, using a Convolutional Neural Network (CNN) to evaluate temporal quality metrics at the PDC level, pinpointing TDSAs while distinguishing them from network disruptions. A new IEEE C37.118 data protocol extension flags TDSA occurrences in PMUs. When activated, the WADC switches to a robust, location-adaptive state observer tailored to the TDSA’s source, designed using Linear Matrix Inequality (LMI) constraints rooted in Lyapunov stability theory. MATLAB simulations across various scenarios confirm the reliability and effectiveness of the proposed methods.