Reinforcement learning based tasks offloading in vehicular edge computing networks

With the rapid development of autonomous and intelligent techniques, vehicles are currently equipped with computation and communication modules for satisfying clients' on-vehicle computing requests. To meet the client's on-vehicle computation requests such as on-vehicle games and self-driving mechanisms, vehicles have to continuously generate computational tasks. However, due to the limited on-vehicle computation capacities, it is barely possible to handle the above requests by the vehicle itself. These requests are then offloaded to special devices such as roadside units or intelligent vehicles. With the fluid feature of the traffic, more requests are generated during the peak hours than at the low hours. Based on the above facts, two significant challenges arise in vehicular edge computing networks: (i) how to accurately determine whether the vehicular networks are in peak or low hours, and (ii) how to effectively offload the generated requests? In this paper, to tackle the above challenges, we investigate the problem of computational requests offloading under different vehicular networking scenarios. To handle the first challenge, we propose the fuzzy inference-based algorithm to identify the situation of the vehicular network (i.e., whether it is in peak hours or low hours). We employ the reinforcement learning-based algorithm for the second challenge to offload the computational requests effectively. Experiments show that our schemes outperform the benchmark by an average of 24.8% regarding resource utilization when satisfying the interests of both service providers and clients.