INTELLIGENT TASK OFFLOADING IN VEHICULAR EDGE COMPUTING NETWORKS

Recently, traditional transportation systems have been gradually evolving to ITS, inspired by both artificial intelligence and wireless communications technologies. The vehicles get smarter and connected, and a variety of intelligent applications have emerged. Meanwhile, the shortage of vehicles' computing capacity makes it insufficient to support a growing number of applications due to their compute-intensive nature. This contradiction restricts the development of ICVs and ITS. Under this background, vehicular edge computing networks (VECNs), which integrate MEC and vehicular networks, have been proposed as a promising network paradigm. By deploying MEC servers at the edge of the network, ICVs' computational burden can be greatly eased via MEC offloading. However, existing task offloading schemes had insufficient consideration of fast-moving ICVs and frequent handover with the rapid changes in communications, computing resources, and so on. Toward this end, we design an intelligent task offloading scheme based on deep Q learning, to cope with such a rapidly changing scene, where software-defined network is introduced to achieve information collection and centralized management of the ICVs and the network. Extensive numerical results and analysis demonstrate that our scheme not only has good adaptability, but also can achieve high performance compared to traditional offloading schemes.