Abstract
Efficient recognition and localization of power quality disturbances (PQDs) are essential for ensuring resilient power distribution. This paper proposes a real-time PQD detection and localization framework for the solar-penetrated IEEE 13-bus system using recurrence plots and deep learning. The network is divided into three zones, with each zone monitored at a selected bus to ensure voltage observability. Three-phase voltage signals are analyzed using a 10-cycle moving window, updated every 250 microseconds, to enable high-resolution disturbance detection. PQD detection is initiated when the cosine similarity index (CSI), computed from the recurrence plot of the moving window of three-phase voltage samples, deviates from that of normal operation and falls below a predefined threshold. This triggers the identification of actual disturbances. Localization is performed using a zone-based detection algorithm that compares the CSI values across all three zones. Real-time signal analysis is conducted on a high-speed x86-based system, while classification is handled on a separate workstation using an EfficientNet model integrated with Squeeze-and-Excitation (SE) blocks. The proposed framework is validated through RTDS simulations and Hardware-in-the-Loop (HIL) testing, demonstrating high accuracy, precise localization, and robustness across various signal-to-noise ratio (SNR) levels.