A Learning-Based Vehicle-Trajectory Generation Method for Vehicular Networking

With the rapid development of mobile applications, networking technologies have been constantly evolved to offer a more convenient way of sharing information and online-communication anytime and anywhere. Vehicular networks have the potential to become one of the important carriers of future mobile networks. The performance of current vehicular networks has been widely evaluated through simulation experiments due to the high cost and impracticality of other experimental approaches. The most paramount factors of vehicle networks are the authenticity of simulative evaluation, where the mobility of the vehicles is the first significant feature (i.e., the nodes of the vehicular network) that must be properly considered. However, generating the corresponding real mobility datasets has always been a big challenge although it is vital to the simulations of vehicular networks. Therefore, in this paper, we propose a learning-based generation method that can be used to build the vehicle-trajectory data for variety of vehicle densities. Firstly, with analyzing the road bayonet data, we obtain the hidden pattern between road traffic and time. Secondly, we deploy Vissim (a well-known traffic simulator) to generate the experimental data by considering the urban functional areas for the origins of vehicles. The generated experimental data are learned by Extreme Learning Machine (ELM), and the weight matrix of the parameters is obtained, which presents the impact of the experimental parameters on the simulation results. We prove the effectiveness of our method by comparing the generated vehicle-trajectory datasets with the vehicle density predicted by the weight matrix and the realistic traffic flow model.