TROVE: A Context-Awareness Trust Model for VANETs Using Reinforcement Learning
TROVE: A Context-Awareness Trust Model for VANETs Using Reinforcement Learning
复制标题
TROVE：使用强化学习的 VANET 上下文感知信任模型
DOI：
10.1109/jiot.2020.2975084
复制
发表时间：
2020-07-01
影响因子：
10.6
通讯作者：
Wu, Dapeng
中科院分区：
文献类型：
--
作者：
Guo, Jingjing;Li, Xinghua;Wu, Dapeng
车联网已经成为一个可见的现实，使车辆之间的信息共享，以提高驾驶安全，并为司机和乘客提供增值服务。然而，由于传感器的缺陷、车辆的恶意等原因，可能会向网络中注入虚假信息，因此，有效的机制来保证车辆使用信息的可靠性在车载网络中具有重要意义。针对这一问题，提出一种上下文感知的信任管理模型，对车辆接收到的信息进行可信度评估，以保证虚假信息不会影响驾驶决策过程。在该方案中，评估请求的信任评估结果由当前情况下的可用相关信息和评估策略决定，不受冲突证据和网络中实体信任水平的影响.此外，我们还设计了一个强化学习模型，允许车辆调整评估策略，以便在不同的驾驶场景中保持准确的评估结果。在不同的驾驶场景进行了大量的实验，以验证所提出的模型的有效性。结果表明，我们的模型是适应不同的驾驶场景，可以忽略不计的时间开销，无论在网络中的恶意节点的比例。此外，与三种不同场景下的最先进的信任模型相比，我们的计划可以实现更高的评估准确率，在非随机道路条件下，没有更多的计算和通信开销。
Vehicular networks have become a visible reality enabling information sharing between vehicles to enhance driving safety and provide value-added services to drivers and passengers. However, false information might be injected into the network because of defective sensors, malicious vehicles, and so on. Therefore, an efficient mechanism to guarantee the reliability of information used by vehicles is of great importance in vehicular networks. To solve this problem, this article proposes a context-awareness trust management model to evaluate the trustworthiness of messages received by vehicles to ensure bogus information will not influence the driving decision-making process. In the proposed scheme, the trust evaluation result of an evaluation request is determined by available related information and the evaluation strategy in the current situation, which is unaffected by the presence of conflicting evidence and the trust level of entities in the network. Moreover, we design a reinforcement learning model that allows vehicles to adjust the evaluation strategy so as to maintain an accurate evaluation result in different driving scenarios. Extensive experiments were conducted in different driving scenarios to verify the effectiveness of the proposed model. The results show that our model is adaptive to different driving scenarios with negligible time overhead, regardless of the proportion of malicious nodes in the network. Furthermore, compared with three types of state-of-the-art trust models in different scenarios, our scheme can achieve a higher evaluation precision rate with no more computational and communication overhead in nonrandom road conditions.