Unsupervised 1D CNN-Based False Data Injection Detection in Smart Microgrid Power Layer Monitoring
Year: 2026
Type: conference-paper
Abstract: Smart microgrids integrate distributed energy resources - solar inverters and wind turbines - that report electrical telemetry (voltage and current) to centralized monitoring systems. These sensor readings are critical for real-time control decisions, making them attractive targets for False Data Injection (FDI) attacks, where an adversary manipulates reported values to mask dangerous operating conditions or destabilize the grid. This paper proposes an unsupervised 1D Convolutional Neural Network (1D CNN) for detecting FDI attacks in smart microgrid power layer telemetry. The model is trained exclusively on normal electrical operating data, learning to forecast normal short-term temporal dynamics. Three FDI attack scenarios are evaluated - replay, scaling, and random injection - and the 1D CNN is compared against a One-Class SVM baseline under identical unsupervised conditions. The framework is integrated into an existing web-based microgrid monitoring platform with a .NET Core backend, Python FastAPI ML service, and React.js dashboard. Across the three attack types the 1D CNN reaches a mean F1 of 0.946 (std 0.011), against 0.823 (std 0.017) for the One-Class SVM baseline, while both sustain a false-positive rate near the 1% calibrated budget.
Authors: Behar Haxhismajli, Galia Marinova
Institution: Technical University of Sofia
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Funder: Technical University of Sofia
OpenAlex ID: W7225720333