Data-driven Vehicular Communications in Urban Vehicular Network

Vehicular communications, referring to data packets exchange among vehicles and infrastructures, have attracted a lot of attention recently because of its contributions in intelligent transportation systems. Due to the complexity of large-scale network topology and dynamic of mobile nodes, vehicular communication is difficult to be established in an ultra-reliable and low-latency way. Existing efforts of vehicular communications mainly focused on historical trajectories and perform experiments on simulated road networks with free moving vehicles, or just analyze a limited scope of traffic pattern over public vehicles. However, previous studies ignored the traffic pattern of urban private vehicles, leading to a narrow perception for data transmission. To make up for this blind spot, we propose a data-driven method for vehicular communications by analyzing the traffic pattern of urban vehicles. We model the urban vehicles traffic pattern by deep neural networks and clustering technique is applied to enhance the ability of collecting data for vehicles. Additionally, to deal with the unbalanced traffic load in road networks, we devise a novel method to detect the appropriate location for deploying auxiliary WiFi-spots to help data transmission in low traffic load area. Therefore, our solution enables the data packets to be transmitted in an optimal way thus comprehensive and valuable information is collected for guiding the transmission. Our proposed method outperforms the state-of-the-art vehicular communication methods in terms of diffusion speed and range in urban vehicular networks.