In this paper, we consider the problem of event localization in the presence of anomalous nodes, in Internet of Things (IoT) and Mobile Crowd Sensing (MCS) systems. A sensing node could be anomalous due to faultiness in any of its components, or due to maliciousness, where it may forge and inject false readings. In both cases, anomalous nodes can significantly alter the task quality and outcome, which may lead to catastrophic consequences, especially in sensitive applications. The current localization systems are not designed to account for the probability of having anomalous readings, hence subjecting them to high errors. Additionally, current anomaly detection systems are not well suited for localization tasks because they are neither dynamic nor continuous, and they do not account for the radial-spreading patterns of data in localization tasks. To overcome these challenges, a Resilient Fault-proof Localization System (RFLS) is proposed, which a) includes an anomaly detection process designed specifically for localization tasks using means of data-based clustering and centroiding, b) dynamically integrates greedy- and genetic-based active nodes selection, Bayesian-based data fusion, and anomaly detection processes in one full localization system, and c) assesses and updates the nodes’ reputations to ensure better performance in future tasks. The efficacy of the proposed system is validated by running experiments for single and sequential localization tasks, for varying conditions, and by using a real-life dataset of the vehicular mobility traces in the city of Cologne, Germany. The results demonstrate that anomalous nodes are efficiently detected, eliminated, and penalized, which in turn greatly improves the accuracy of the localization tasks.

在本文中，我们考虑了在物联网（IoT）和移动人群感知（MCS）系统中存在异常节点的事件定位问题。感测节点可能由于其任何组件的故障或由于恶意而异常，在此可能伪造并注入错误的读数。在这两种情况下，异常节点都会显着改变任务质量和结果，这可能导致灾难性后果，尤其是在敏感应用程序中。当前的本地化系统没有被设计为考虑具有异常读数的可能性，因此使它们遭受高误差。另外，由于当前的异常检测系统既不是动态的也不是连续的，并且它们不考虑定位任务中数据的径向扩展模式，因此它们不适用于定位任务。为克服这些挑战，提出了一种弹性防故障定位系统（RFLS），其中a）包括一个异常检测过程，该过程专门为基于数据的聚类和质心化的定位任务而设计，b）动态集成了贪婪和遗传-一个完整的本地化系统中基于活动节点的选择，基于贝叶斯的数据融合和异常检测过程，以及c）评估和更新节点的信誉，以确保在未来任务中表现更好。通过运行针对单个和顺序的本地化任务，针对不同条件的实验，以及通过使用德国科隆市的车辆通行轨迹的真实数据集，可以验证所提出系统的有效性。结果表明，异常节点得到了有效的检测，消除和惩罚，