Detection and mitigation of False Data Injection Attacks in vehicle platooning using Kalman Filters and Penalized Weighted Least Square

Abstract

Vehicle platooning provides significant improvements in traffic efficiency, fuel economy, and road safety, but it is highly vulnerable to cyber–physical threats such as False Data Injection Attacks (FDIAs). These attacks can compromise string stability, disrupt coordination, and threaten the safe operation of the platoon. To address these challenges, this dissertation proposes a robust framework for the detection, identification, and mitigation of malicious data manipulations in vehicle platoons. Two complementary methods are developed and analyzed: (i) a Kalman Filter–based detection scheme for platoons with constant (zero) lead-vehicle acceleration, and (ii) a combined Unknown Input Kalman Filter (UIKF) and Penalized Weighted Least Squares (PWLS) estimator for platoons with variable lead-vehicle acceleration. In the first method, the standard Kalman Filter is employed to estimate vehicle states such as position, velocity, and acceleration when the lead vehicle maintains a constant acceleration. The residuals between the estimated and measured states are continuously monitored to identify abnormal deviations caused by injected false data. Once a residual exceeds a predefined threshold, an attack is detected and its source is localized. This approach provides fast and accurate detection in steady-state operating conditions, ensuring stability of the platoon when external disturbances are minimal. The second method extends the analysis to more realistic cases where the lead vehicle experiences time-varying acceleration. Here, an Unknown Input Kalman Filter (UIKF) is designed to estimate both the vehicle states and the unknown input representing the leader’s acceleration, effectively removing uncertainty about its dynamics. The estimated states are then refined through a Penalized Weighted Least Squares (PWLS) estimator, which assigns lower weights to suspicious or inconsistent measurements. This adaptive weighting suppresses the influence of corrupted data and enhances resilience against noise and coordinated attacks, allowing accurate reconstruction of the true vehicle states even under severe disturbances. The effectiveness of both methods is validated through extensive Python simulations under various operating conditions, including random, stealthy, and coordinated multivehicle attacks. Simulation scenarios cover both constant- and variable-acceleration platoons, demonstrating high detection accuracy (average 98.62%), low false-alarm rates, and accurate recovery of attacked state variables. Comparative analyses with conventional observer-based methods highlight superior detection performance and computational feasibility for real-time deployment. By integrating both fixed- and variable-acceleration models, this thesis provides a unified, noise-resilient, and computationally efficient framework for securing vehicle platoons against FDIAs. It ensures reliable detection and mitigation across a wide range of cyber–physical threat scenarios, preserving safe, coordinated, and energy-efficient platoon operation.