[2609.17635] Physics-Constrained Digital Twins for Sensor Integrity in Urban Pedestrian Flow: Detecting Stealthy False Data Injection with Conformal Guarantees

Physics-Constrained Digital Twins protect urban pedestrian-flow sensors by embedding flow conservation laws into the twin’s architecture, which shrinks the space available for stealthy false data injection attacks.

• Attack Margin: On six years of Melbourne data, the physics constraint reduced worst-case flow corruption by 0.54 with one compromised device and 0.19 when a third of the fleet was compromised.
• Detection Method: The system combines a conservation residual with innovation scores, using adaptive conformal calibration to set alarm thresholds that handle seasonal drift and maintain guaranteed false-alarm rates.
• Key Finding: The security gain derives from the conservation law rather than spatial locality; replacing the street graph with a distance graph collapses the attack margin to 0.09.
• Performance: The approach detects coordinated injections three times faster than strong published detectors at the same false-alarm rate, with a single compromised device causing only 4.3% misattribution of pedestrian-hours (down from 9.4% without constraints).

This work addresses sensor-integrity risks in urban pedestrian-flow monitoring, where sensors such as counters, cameras, LiDAR, or beacon systems feed data into traffic management, safety planning, and evacuation systems. The paper proposes physics-constrained digital twins as a way to detect stealthy false-data-injection attacks: adversarial perturbations that alter sensor readings while remaining superficially plausible and consistent with normal pedestrian dynamics. Rather than relying only on statistical outlier detection, the approach embeds domain-specific physical constraints—such as flow conservation, capacity limits, spatial–temporal consistency, and boundary conditions—into a digital twin that predicts how pedestrian flow should evolve under plausible operating regimes.

A key contribution is the use of the digital twin as an online integrity verifier. Sensor observations are compared against physics-consistent model predictions, and discrepancies are analyzed to identify injections that may evade purely data-driven detectors. The paper further incorporates conformal prediction to calibrate detection thresholds in a statistically principled way, aiming to provide finite-sample guarantees on false-alarm or miscoverage behavior even under distribution shift, noise, or imperfect model calibration. This is especially relevant for stealthy attacks, which are designed to stay within the envelope of “normal” behavior while still biasing downstream estimates or decisions.

The significance of the work lies in connecting cyber-physical security with pedestrian-mobility modeling. It moves digital twins from being predictive or planning tools toward serving as auditable integrity monitors for safety-critical urban infrastructure. By combining physics-based modeling with conformal guarantees, the approach offers a more defensible alternative to heuristic anomaly detection, potentially supporting trustworthy control, incident response, and risk assessment in pedestrian-heavy environments.