Detection And Correction Method For Abnormal Data Over Data Streams Of Sensor Networks

It is the key technology to detecte and correct abnormal data in the sensor networks application. With the quick developing of data streams dealing technique and the data fusion technique the detection and correction for abnormal data will have widely developing spacing.In this thesis, the recent method for detection and correction for abnormal data over data streams of sensor networks is analysed firstly, and a spatial-temporal architecture model for detection and correction for abnormal data was proposed according to the deficiency of exist methods. The temporal denoising method based on wavelet transforms and the spatial data fusion method based on BP neural network are combined in the model. Based on this model, a detection and correction for abnormal data algorithm based on wavelet scale is proposed. The time window used in data fusing is determined by the time threshold in the algorithm. Using the time division and frequency division characters of wavelet transform, detection and correction for abnormal data is related to the sustaining time of abnormal data. The wrong report of abnormal data is gotten rid of by the correcting of the sensor networks abnormal data using the data fusing result of muti-sensors.By using the wavelet transforms, the data stream can be compressed and denoised in the mothed. The data transmitted between the sensors can be reduced so as to the using time of the sensors can be increased by saving energy consumption. The affection of noise to the fusing result can also be reduced by using the method. The times of fusion is reduced by the data fusing algorithm based on the wavelet scale, so the mothed has more better time efficiency and makes the detection and correction for abnormal data more reliable in the sensor networks.