Dynamic Source-Aware Transfer Learning for Vehicular Platoon Predictive Control

The allocation of computational resources in the vehicular platoon is susceptible to communication topology, particularly under complex scenarios, such as off-ramps. This article proposes a dynamic source-aware transferable data-driven control (DSTDC) strategy to optimize computing resource allocation under the off-ramps scenario. A data-driven predictor is established to characterize the powertrain structure of vehicular platoon. To improve the computational efficiency of platoon control algorithm, a transfer learning method is constructed. Specifically, the knowledge of vehicle in the source domain is transferred to vehicles in the target domain, and the latter just should update the knowledge in terms of the individual dynamic characteristics and tasks. Furthermore, an adaptive domain switching mechanism is developed to address the issue of source domain vehicle departure in the off-ramps scenario. Here, the source domain vehicle will be subject to reevaluation in terms of the computing performance, data accumulation and driving stability. Finally, a platoon experiment platform is conducted, and the results demonstrate the effectiveness of the proposed DSTDC strategy.