Privacy-Aware Data-Driven Predictive Control for Mixed Traffic Via Masking Techniques

Data-driven predictive control has gained attention as a model-free alternative, particularly when obtaining an explicit system model is challenging or costly. However, its application to mixed traffic systems requires data sharing, raising significant concerns about privacy leakage. This paper presents a novel masking-based strategy to enable privacypreserving data-driven predictive control in mixed traffic environments, where connected and automated vehicles (CAVs) operate alongside human-driven vehicles (HDVs). The proposed method protects against external eavesdroppers who intercept communication channels in an attempt to infer the states and control inputs of CAVs. By integrating an affine transformation with random noise injection, our strategy obfuscates actual system states and inputs, leading to a modified data-enabled predictive control framework that ensures both privacy protection and optimal control. We demonstrate that this method can achieve differential privacy without degrading control performance or imposing heavy computational burdens. Simulation results validate the effectiveness of the proposed approach.