Research On Clustering And Kronecker Based Optimized Sensory Data Compression

In recent years,WSN technology is developing with the breathtaking speed,the demand for intelligent manufacturing,intelligent factory and intelligent service is also exploding.Hence,industrial IoT(IIoT)arises at the historic moment and becomes a promising field which supports industrial applications,improves manufacturing and service efficiency by intelligent decision making.At present,industrial sensors and IoT devices are rapidly and continuously deployed in the industrial environment.IoT plays an increasingly important role in intelligent industrial applications.Although IoT brings much convenience,it also brings new challenges caused by the large amount of data generated by industrial sensors.Various industrial devices put forward strict requirements for data collection,processing and transmission.Therefore,it becomes a huge challenge at present that how to efficiently collect,process and transmit industrial big data in the IIoT scenario.Aiming at the existing prominent problems existing in the current sensing data compression scheme on clustering mechanism researching,spatio-temporal correlation exploring,measurement matrix optimizing,dictionary training and other aspects,this thesis operates the research on clustering and Kronecker based optimized sensory data compression in the industrial Internet of things scenario to solve the above problems.The main innovations include the following three aspects:1)This thesis proposes a k-means based Kronecker spatio-temporal(two-dimensional)compression mechanism.Firstly,this mechanism constructs a k-means clustering algorithm,which is used to deeply and effectively explore the spatial correlation in sensory data.It also provides support for obtaining better data compression effect.Secondly,a novel Kronecker spatio-temporal compression method is constructed,which combines the traditional spatio-temporal compression with Kronecker product.This method not only guarantees the successful execution of CS reconstruction algorithm,but also better integrates and utilizes the spatio-temporal correlation.This method effectively improves the reconstruction accuracy of sensory data.2)This thesis proposes a fog computing based Kronecker compression mechanism with measurenment matrix optimization.In view of the performance improvement for the above proposed mechanism,this part makes the extensive research.This part introduces the measurement matrix optimization based on Kronecker spatio-temporal compression.And the entire compression mechanism is flexibly applied to the IIoT scenario,which not only guarantees the successful execution of CS data reconstruction,but also greatly reduces the redundancy data transmission between the fog and cloud nodes by data pre-processing operation in the fog node.In addition,the established optimized model of Kronecker concatenated measurement matrix designs an optimal solving method for Kronecker spatio-temporal concatenated measurement matrix.The optimization method can minimize the mutual correlation between the measurement matrix with sparse basis and reduce the self-correlation of the measurement matrix,which make the matrix adapt to the real scene and achieve the goal of further improving the data reconstruction quality.3)This thesis proposes a high quality data compression mechanisms by integrating dictionary training and measurement matrix optimization.The above two mechanisms only consider the research on clustering and optimization,which don't consider the dictionary training and the relationship between the measurement matrix with sparse basis.This part conducts the deep exploration on the correlation between the measurement matrix and sparse basis through joint optimization of sparse dictionary and measurement matrix.Firstly,this mechanism constructs a two-dimensional dictionary training method based on K-SVD to realize the sparse representation of IIoT spatio-temporal data.Secondly,the column coherence of the sensing matrix is minimized by optimizing the corresponding measurement matrix in the spatial and temporal domains.Finally,a joint optimization method is proposed,which can perform the tradeoff between dictionary training and optimization of measurement matrix.So that the reconstruction error can reach the theoretical minimum.