While sensor networks have been used in various applications because of the automatic sensing capability and ad-hoc organization of sensor nodes, the fault-prone characteristic of sensor networks has challenged the ev...While sensor networks have been used in various applications because of the automatic sensing capability and ad-hoc organization of sensor nodes, the fault-prone characteristic of sensor networks has challenged the event detection and the anomaly detection which, to some extent, have neglected the importance of discriminating events and errors. Considering data uncertainty, in this article, we present the problem of data discrimination in fault-prone sensor networks, analyze the similarities and the differences between events and errors, and design a multi-level systematic discrimination framework. In each step, the framework filters erroneous data from the raw data and marks potential event samples for the next-step processing. The raw data set D is finally partitioned into three subsets, Devent, Derror and Dordinary. Both the scenario-based simulations and the experiments on real-sensed data are carried out. The statistical results of various discrimination metrics demonstrate high distinction ratio as well as the robustness in different cases of the network.展开更多
文摘While sensor networks have been used in various applications because of the automatic sensing capability and ad-hoc organization of sensor nodes, the fault-prone characteristic of sensor networks has challenged the event detection and the anomaly detection which, to some extent, have neglected the importance of discriminating events and errors. Considering data uncertainty, in this article, we present the problem of data discrimination in fault-prone sensor networks, analyze the similarities and the differences between events and errors, and design a multi-level systematic discrimination framework. In each step, the framework filters erroneous data from the raw data and marks potential event samples for the next-step processing. The raw data set D is finally partitioned into three subsets, Devent, Derror and Dordinary. Both the scenario-based simulations and the experiments on real-sensed data are carried out. The statistical results of various discrimination metrics demonstrate high distinction ratio as well as the robustness in different cases of the network.
