Research On Incomplete Data Stream Imputation Method For Internet Of Things

With the rapid development of Artificial Intelligence and the Internet of Things(Io T),data-based intelligent services have been promoted to multiple application fields.However,due to the complexity of the Io T environment,the sensor may produce incomplete data streams in data collection and transmission.Incomplete data streams will destroy the integrity of the dataset,and then affect the accuracy of model analysis.In addition,for some intelligent Io T services that need real-time data streams,the existence of incomplete data streams will paralyze these services.Therefore,incomplete data streams imputation in the Io T environment is an urgent problem to be solved.This paper mainly studies the imputation of incomplete data streams in the Io Ts environment.The main contents of this paper are as follows:(1)An incomplete data streams imputation method based on Ra-LSTM is proposed for offline incomplete data streams in the Io Ts.The imputation method for offline environment needs comprehensively consider the data imputation accuracy and imputation efficiency.Although the existing imputation method based on statistics is simple and feasible,it has the problem of poor data imputation accuracy.Most machine learning and deep learning methods have the problem of low imputation efficiency.To solve the above problems,this paper proposes a relationship-aware LSTM imputation method for offline incomplete data streams.In this method,the temporal LSTM(T-LSTM)model and spatial relational LSTM(SR-LSTM)model are used to extract the temporal and spatial features of the sensor data,and then the spatio-temporal fusion algorithm was used to fuse the temporal and spatial features,and the fused features were used to impute the missing data in the incomplete data streams.Finally,the Ra-LSTM imputation method is verified by experiments on three Io T sensor datasets in different fields.(2)A fast incomplete data streams imputation method combining Similarity Reduction technology and Matrix Factorization theory(SRa MF)is proposed for online imputation of incomplete data streams.The imputation method for online environment mainly considers the imputation real time.Some existing imputation methods based on statistical indicators have good real time but poor imputation accuracy.To solve the above problem,this paper combines similarity reduction technology and the theory of matrix decomposition,proposes a fast incomplete data streams imputation method,the thought of SRa MF is the initial sensor data matrix preprocessing,the original dataset for data reduction,the use of residual reduction of incomplete data streams for rapid imputation.The experimental results show that SRa MF imputation method can improve the time performance by more than 80% on the premise of ensuring the imputation accuracy.SRa MF method reduces data redundancy and improves imputation real-time on the premise of controllable imputation accuracy.SRa MF can provide complete data streams for downstream intelligent Io T services in real time,so that intelligent services can make real-time analysis and quick decision.