Скачать или смотреть MLP vs Ensembles: The Real-Time Power Grid Defense Secret

⚡ *AI-Powered Grid Protection: The Real Test of Machine Learning*

The grid is changing — solar, wind, and batteries replace traditional generators, making fault detection harder. Inverters limit fault current and create noisy waveforms, while cyberattacks can mimic real faults. Can AI tell the difference in real time?

In this deep dive, we reveal the truth: **offline accuracy is a trap**. We tested 12 models in a high-fidelity, streaming environment using WinIGS at 4.8 kHz (80 samples/cycle) — simulating real inverter dynamics, timing drift, and 18 real-world anomalies, including faults, CT/PT attacks, and GPS spoofing.

Results? *MLPs outperformed ensembles* — sustaining 98%+ classification coverage, rejecting false alarms, and adapting to transients. Ensembles prioritized precision but abstained too often — dangerous in protection.

Key insight: **Abstention kills reliability**. In real time, you can’t afford to wait.

We open-sourced the full dataset & code: [GitHub.com/Emadeddin/T4Tech](https://github.com/Emadeddin/T4Tech)

👉 LIKE, SUBSCRIBE, and comment: Should AI replace traditional relays?

#AIinPowerSystems #SmartGrid #MachineLearning #PowerGridProtection #InverterBasedResources #CyberPhysicalSecurity #DeepLearning #GridResilience #RealTimeAI #MLforGrids
Read more on arxiv by searching for this paper: 2511.06714v1.pdf