Research On Reconstruction Method Of Sensor Sensing Loss Data Based On Spatio-temporal Correlation

With the development of information technology,various automatic and intelligent management and control technologies have been widely used.as an important part of informatization,Internet of Things(IoT)has become one of the key research issues.Sensor data is very important for the application of the IoT.One of the values of the IoT is to provide useful and reliable data for the upper application.Due to the characteristics of the sensor structure and some external influences,the lack of sensor data is inevitable.In order to better mine effective data information,it is particularly necessary to estimate missing data.At present,most data reconstruction methods are based on single attribute,which has certain limitations.This article establishes a model which can estimate missing data more accurately by analyzing the characteristics,spatial-temporal correlation of each attribute.Firstly,the characteristics of perceptual data are analyzed through real data sets.In this article,we analyze the temporal stability and spatial correlation of the data,describe and analyze the data of the same node at different times and the relationship between the data perceived by different nodes at the same time through data visualization.Secondly,a data missing estimation and reconstruction model is established.On the basis of the above correlation analysis,this article establishes a multiple linear regression LR model based on time stability,a BP neural network model based on spatial correlation,and the parameters of the model are determined.Then the missing data are estimated separately.Then,the corresponding weights are given to the estimated results through the Softmax function,and Establish a reconstruction model based on time-space correlation.Finally,an example is given to validate the model based on data missing reconstruction.In this paper,MSE is used as the performance evaluation criterion.Several models proposed in this article are simulated on real data sets.By adjusting different parameters,the prediction results of different models are evaluated and analyzed.The results show that the LR_BP model based on temporal-spatial correlation is more accurate than the LR_BP model based on single attribute.By comparing with LIN and KNN,it is found that the linear regression method and BP neural network method used in this article have better reconstruction effectand smaller error than other methods.