In modern power systems, secure operation depends on the integrity of measurement and control data. When sensor streams or telemetry are compromised, through false data injection, replay, or coordinated perturbations, the Energy Management System (EMS) can be driven to compute dispatch and control actions from an incorrect estimate of the grid state. Because these manipulations can be structured to remain statistically plausible, they may evade simple thresholding or residual-based checks, increasing the risk of constraint violations and degraded reliability.
That’s exactly the weak point Hemmati & Saboori go after in their paper “A hybrid data-driven and physics-informed energy management model for electrical grids with spatio-temporal cyberattack detection–reconstruction using digital twin,” published in Energy Conversion and Management (DOI: 10.1016/j.enconman.2025.120728). What makes their approach compelling is that it treats cybersecurity and energy optimization as a single operational loop. The authors model a full 24-hour cycle: the grid state is first captured through a real-time updated nonlinear digital twin, then transmitted via typical operational channels, and continuously checked for attacks. When the system detects malicious manipulation, it doesn’t stop at raising an alarm, but it reconstructs the corrupted data so the EMS can keep running on a restored, physically plausible picture of the grid.
This is where the paper aligns with the AI-DAPT project’s core thesis, and at large with its Energy demonstrator: reliable AI in the real world requires hybrid intelligence combined with lifecycle discipline. AI-DAPT is not just interested in “accurate models”. Rather, it targets an AI-Ops / intelligent pipeline lifecycle that improves systems continuously, while keeping human-in-the-loop oversight and combining data-driven learning with first-principles reasoning for trust and robustness.
Hemmati & Saboori essentially provide a blueprint for what “AI-DAPT-ready” energy intelligence looks like in practice: a hybrid model grounded in physics, wrapped in observability, closing the loop between detection, diagnosis and recovery, and designed for operations where data can drift, sensors can fail, and adversaries can actively deceive. It’s the same design philosophy as the AI-DAPT hybrid-model approach of AI becoming a resilient co-pilot, auditable, monitorable, and recoverable, when the grid is under pressure.
