Abstract
One of the important sources for big data is the data sets collected by wireless sensor networks. However, errors in sensor data could result in serious damages in industrial applications. Therefore, error detection plays a crucial role in industrial wireless sensor networks (IWSNs). Existing approaches of error detection are generally threshold-based, which rely on a predetermined threshold to judge whether a reading is erroneous. The threshold-based approaches, however, often fail to balance between detection accuracy and false alarm rate. It is thus difficult, if not impossible, to obtain a proper threshold for various errors in real-world applications. Motivated by this consideration, we propose a novel Threshold-free Error Detection (TED) approach for IWSNs. By taking the advantage of the spatio-temporal correlations between sensor readings, we present the model to character the relationship between sensor pair and thus construct a correlation graph for IWSN. In the correlation graph, the states of nodes, i.e., the states of sensor readings, are accurately exploited without requiring any threshold. Through the experiments on both real sensors and simulations, we demonstrate that the proposed approach is able to significantly improve the detection accuracy while largely reduce the false alarm rate.
Funding Information
  • National Natural Science Foundation of China (61420106009, 61402542, 61672543)

This publication has 26 references indexed in Scilit: