Self Learning-Based Platooning Control Strategy for Connected Autonomous Vehicles With Switching Topologies

Connected autonomous vehicles (CAVs) cooperate through vehicle to every-thing (V2X) wireless communication to achieve platooning control, which can substantially increase the capacity of roads and reduce the risk of accidents. However, wireless communication systems may encounter abnormal situations in actual traffic environments, such as data packet loss, signal interruption, and communication module damage, which will cause the switching of communication topology among vehicles, thus affecting the performance of platooning system. To address this problem, a self learning-based comfort and energy-efficient platooning control strategy through cloud for CAVs with switching topologies and state estimation is proposed. First, we use weighted data fusion to process the original data. Simultaneously designing a rule to ensure that the acceleration change rate at switching moment remains smooth. Then, we introduce the adaptive Kalman filter (AKF) to get a high-precision state estimation in the abnormal communication scenario and thus to ensure acceptable passenger comfort level. Finally, a double-dueling-deterministic policy gradient ( $\rm \mathbf {D^{3}PG}$ ) algorithm is proposed to achieve the platooning control, and a reliable self-learning platoon strategy was established using iterative learning, prior knowledge and data correction. The numerical results indicate that the proposed method can ensure car-following performance, comfort, and energy efficiency of CAV in abnormal communication scenarios.