VANET云环境下基于人工神经网络的车辆任务卸载策略

Artificial neural network methods for task offloading in VANET cloud

在VANET(vehicular ad-hoc network)云环境中,由于车辆自身资源受限等原因需要将部分计算密集型任务卸载至周围车辆协同处理。而车辆移动的随机性是影响车辆任务卸载性能好坏的重要因素之一,对此,提出了一种车辆间的卸载任务分配策略。考虑到车辆之间连接时间的随机性,提出一种基于人工神经网络的连接时间预测方法,该方法能够通过对历史数据的学习,较为准确地对未来车辆行驶轨迹进行预测。此外,车辆将空闲资源进行共享意味着自身能耗增加,由于车辆本身的自私性使得车辆不会无偿为周围车辆提供服务。为了激励车辆之间进行协作,制定了一种分布式买卖博弈方法达到车辆资源需求与收益之间的平衡,还设计了一种集中式任务分配策略以获得任务卸载的最大效用。仿真显示,提出的方法在最大化卸载效用与提高任务卸载成功率方面都有较好的性能。

In the VANET cloud,some computationally intensive tasks need to be offloaded to the surrounding vehicles for coordinated processing due to the limited resources of the vehicle itself.The randomness of vehicle movement is one of the important factors affecting the offloading performance.For this reason,we propose an offloading task allocation strategy between vehicles.Considering the randomness of connection time between vehicles,we adopt a connection time prediction method based on artificial neural network.This method can accurately predict the future vehicle trajectory by learning historical data.In addition,sharing idle resources to other vehicles means increasing energy consumption.Due to the selfishness of the vehicles,they will not provide services for surrounding vehicles without compensation.In order to encourage the cooperation of available vehicles,we have developed a distributed Buyer/Seller game method to achieve a balance between resource demand and payment.Finally,we design a centralized task allocation strategy to get the maximum utility of task offloading.The simulation results show that the proposed method is feasible in maximizing the offloading utility and improving the success rate of task offloading.