Abnormal Data Detection and Correction of Sensor Network Data Flow

It proposes a method suitable for sensor network application data flow abnormal time-space comprehensive detection and correction. It combines the wavelet-based time domain filtering method and BP neural network-based time domain data fusion, and based on this, it proposes an abnormal detection and correction method based on wavelet scale, determining the adopted time window of data fusion through the time threshold. Using the time and frequency characteristics of wavelet transform, it is related with abnormal detection and correction and abnormal data lasting time, conducting correction of the abnormal data of sensor network by the multi-sensor data fusion result, so as to eliminate the abnormal data of sensor network. In the sensor network application, it often needs to conduct detection and correction when producing abnormal data. But because of complex sensor application environment, the abnormal detection and correction process will often be interferred by some noise. And because in the sensor network, it often uses multi-sensors to determine in the same district, so to determine whether the measurement data in the district is abnormal, people should not only consider the influence of various sensor data of various noise, but also should use the data result of noise-removing fusion as the sensor network estimation value in the district. In addition, because of numerous network sensor quantity and huge data amount, conducting data fusion of the sampling data of all times will produce larger time-delaying, which is also not good for data flow abnormal detection application. So, people should try to reduce the fusion times. Thus, this paper proposes a method suitable for sensor network application data flow abnormal time-space comprehensive detection and correction.