Federated Learning (FL) provides an effective framework for efficient processing in vehicular edge computing. However, the dynamic and uncertain communication environment, along with the performance variations of vehicular devices, affect the distribution and uploading processes of model parameters. In FL-assisted Internet of Vehicles (IoV) scenarios, challenges such as data heterogeneity, limited device resources, and unstable communication environments become increasingly prominent. These issues necessitate intelligent vehicle selection schemes to enhance training efficiency. Given this context, we propose a new scenario involving FL-assisted IoV systems under dynamic and uncertain communication conditions, and develop a dynamic interval multi-objective optimization algorithm to jointly optimize various factors including training experiments, system energy consumption, and bandwidth utilization to meet multi-criteria resource optimization requirements. For the problem at hand, we design a dynamic interval multi-objective optimization algorithm based on interval overlap detection. Simulation results demonstrate that our method outperforms other solutions in terms of accuracy, training cost, and server utilization. It effectively enhances training efficiency under wireless channel environments while rationally utilizing bandwidth resources, thus possessing significant scientific value and application potential in the field of IoV.