Research On Key Technologies Of Missing Data Imputation In Wireless Sensor Networks

Nowadays,the wireless sensor networks(WSNs)have been used in different areas.However,due to the network structure and the features of the sensor nodes,it’s inevitable for data to get lost in the acquiring procedure which will make the dataset incomplete.It’s the target of applications of WSNs that extract valuable information from sensor data,but most of the tools utilized in the applications,including the data mining and machine learning,only work on the complete dataset without missing data.Currently,to make estimate values of the missing data and replace them is the major way to convert the incomplete data to the complete data.Meanwhile,it is also the most important step in the preprocessing of sensor data.Therefore,the imputing algorithms of missing data have become one of the important research areas of WSNs.In this dissertation,we focus on the key technologies to tackle problems of missing data imputation and the research includes the typical missing data imputation in WSNs and big data based missing data imputation.The main results are described as follows:(1)To solve the problem of both improving the accuracy and the successful imputation rate of the offline missing data imputation in WSNs,from the perspective of the temporal and spatial distance measurement,a temporal and spatial nearest neighbor values based offline missing data imputation algorithm is proposed and it fully exploits correlation of the temporal and spatial nearest neighbor values to make the imputation.Compared with current temporal and spatial correlation-based imputation algorithms,the experiment results show it improves both the accuracy of imputation and the successful imputation rate.(2)To solve the problem of improving the accuracy of the online missing data imputation in WSNs with reasonable time of imputation,a virtual temporal neighbor based online missing data imputation algorithm is proposed.It fully utilizes the correlation of sensor data stream to make the imputation by the virtual temporal neighbors.Compared with current online imputation algorithms,the experiment results show it improves the accuracy of online imputation of sensor data stream with relatively reasonable time of imputation for WSNs applications.(3)To solve the problem of implementing missing data imputation of big sensor data based on its features,a data decomposing based missing data imputation algorithm for big sensor data is proposed.It exploits the correlation in big sensor data and decomposes the data by three vector dimensions before deploying the missing data imputation tasks on the parallel computing platform.Compared with serial imputation algorithms,the experiment results show it ensures the accuracy of imputation with reduced time of imputation.(4)To solve the problem of implementing the accurate approximate querying application on big sensor data with missing values,a double sampling based missing data imputation algorithm for big sensor data is proposed.It deals with the querying application for the average value of big sensor data by making double sampling on big sensor data and making missing data imputation on the small sample data then making the querying.As results,it returns the approximate querying result on the big sensor data with confidence intervals.The experiment results show it ensures the accuracy and the credibility of approximate querying for the average value with reasonable querying time.