As power grids transition toward inverter-based resources—solar, wind, and battery storage—traditional protection schemes face new vulnerabilities. Inverters respond to electrical disturbances differently than synchronous generators, and digital measurement systems introduce cyber-physical attack surfaces that relays were never designed to guard against. Researchers at NREL and partner institutions have benchmarked a novel machine learning approach to address these gaps.
The study compares two protection strategies on identical high-speed measurements from a simulated grid rich in inverter resources. The first, Dynamic State Estimation-Based Protection (DSE-EBP), acts as a fast anomaly detector, flagging abnormal conditions in 0.4 to 1.7 milliseconds. The second, DL-Xformer, is a Transformer neural network that classifies what kind of event occurred among 18 scenarios: normal operation, 11 physical faults (short circuits, line breaks, etc.), and six cyber-physical attacks including GPS spoofing and current transformer tampering.
Results show DL-Xformer correctly identified fault and attack types in 2.5 to 50 milliseconds, with typical classification taking about 13 milliseconds. Even under extreme stress—when a cyberattack coincides with lingering transients from a previous fault—the network ultimately converged to the correct diagnosis. The model's decisions were grounded in physically meaningful current and voltage signatures, making the AI interpretable to engineers rather than a black box.
The research proposes a layered defense architecture: DSE-EBP provides immediate warning, giving control systems time to act, while DL-Xformer runs in parallel to classify the threat type and inform targeted response strategies. As inverter penetration climbs toward 50 percent or higher in many regions, such adaptive protection becomes essential. The work demonstrates that modern AI and traditional grid science can integrate effectively, offering vendors and utilities a path toward smarter, more resilient protection for the renewable energy era.
