AI-Driven Adaptive Protection Schemes for Inverter-Dominated Smart Power Grids with High Renewable Energy Penetration

Falguni Shah, Hetal Modi, Anurag Patel

Abstract


The ongoing global transition toward sustainable energy paradigms has catalyzed an unprecedented integration of renewable energy sources (RES), such as solar photovoltaics (PV) and wind power generation, into modern power transmission and distribution grids. Consequently, power systems are rapidly evolving from conventional synchronous machine-dominated networks into inverter-dominated smart power grids. While inverter-based resources (IBRs) provide essential decarbonization benefits, their fault current contributions are strictly limited by semiconductor device thermal constraints—typically capped at 1.1 to 1.5 times nominal rated current. This radical reduction in fault current magnitudes, combined with non-standard fault current signatures, low or zero system physical inertia, and bi directional dynamic power flows, severely degrades the sensitivity, selectivity, speed, and overall reliability of legacy overcurrent and distance protection relays. To address these critical protection vulnerabilities, this paper presents a novel AI-driven adaptive protection scheme designed specifically for inverter-dominated smart power grids with high renewable energy penetration. The proposed framework employs a decentralized machine learning architecture leveraging quantized Temporal Convolutional Networks (TCNs) and dynamic adaptive thresholding. High-frequency electrical waveforms (voltages and currents) are continuously sampled from local Intelligent Electronic Devices (IEDs) across microgrids and transmission feeders. The model extracts localized transient, harmonic, and phase-angle features to instantaneously classify fault types, estimate fault locations, and adaptively modify relay trip thresholds in real time without relying on continuous wide-area high-bandwidth communication infrastructure. The framework was comprehensively evaluated using high-fidelity Real-Time Digital Simulator (RTDS) and MATLAB/Simulink benchmark systems across IEEE 39-bus and 13-bus microgrid testbeds under varying levels of IBR penetration ranging from 30% to 90%. Experimental results demonstrate that the proposed AI-driven protection scheme achieves a fault classification and detection accuracy of 99.4%, reduces relay tripping latency to under 12.5 milliseconds, and completely mitigates relay blinding, sympathetic tripping, and failure-to-trip events under grid-connected and islanded operating modes. Furthermore, the integration of lightweight edge-computing algorithms minimizes computational overhead, ensuring execution times below 8 milliseconds on standard digital protection relays. This study establishes a robust, resilient, and scalable protection framework essential for enabling high penetration levels of clean energy while maintaining smart grid stability.

KEYWORDS: Adaptive Protection, Inverter-Dominated Grids, Renewable Energy Penetration, Inverter-Based Resources (IBRs), Temporal Convolutional Networks, Smart Grid Resilience, Relay Blinding, Machine Learning.

Full Text:

PDF 67-81

Refbacks

  • There are currently no refbacks.