| With the rapid development of information technologies,computer technologies and communication technologies,human living environment is becoming a hyper space in-volving the cyber space,physical space and social space,which is also recalled as"Cyber-Physical-Social Systems"(CPSS).Currently,how to provide perspective and personal ser-vice for human is one of the main research issues of CPSS.However,data,flowing around the three spaces,record the behavior trace of users,hiding the requirement,interesting and habit of humans.Therefore,data are selected as the starting point of research.However,data collected from the CPSS have the main features as following.Firstly,collected from independent devices and multiple-sources,data are influenced by the char-acteristics of the generating devices and heterogeneous.Secondly,data are large scale and rapid produced.Billions of gigabytes of data are generated every second about all aspects of our daily life in CPSS.Thirdly,collected data are noisy and redundancy,which are needed to be processed.Therefore,how can we obtain the high quality data set of the collected data?And how can we match users’ requirement,interesting and habit under different attributes conditions?Based on the previous work of our research team that tensor is used to represent the big data,the main contributions of this thesis are summarized as following,Firstly,a general systematic framework,namely "Big Data-as-a-Services" framework including the sensing plane,cloud plane and application plane is processed for CPSS ser-vices,according to wide distribution,large scale collected data and widely application of the CPSS.In the sensing plane,the local tensor is used to present the local CPSS,which will be initially processed.In the cloud plane,global tensor,describing the global CPSS,will be obtained by integrating the local tensors together.Also,the global tensor will be decomposed to obtain the high quality data set,according to the tensor decomposition theo-ry.Then,the high quality data set can be used to match users’ requirement,interesting and habit.Secondly,to obtain the high-quality data set,the distributed and incremental comput-ing methods are proposed.Once the representation is decided,the high-order singular value decomposition is used to realize the tensor decomposition.According to the tensor di-vided or incremented along a certain order,the distributed decomposition method with its incremental computing method is proposed,after analyzing the challenging questions in its unfolding matrices.To realize the tensor divided or incremented along several or even all orders at same time,a tree-based multi-order distributed high-order singular value decom-position with its incremental computing method is processed,after tensor blocks unfolding integration regulation proposed to analyze the unfolding matrices integration.Also,the tree-based multi-order distributed high-order singular value decomposition with it incre-mental computing method has several disadvantages such as nodes in high layer of tree with heavy workload,low degree of parallelization.Accordingly,a tree-based ring distribut-ed high-order singular value decomposition with its incremental computing method and the tree-based tree distributed high-order singular value decomposition with its incremental computing method were proposed to address these mentioned challenges in this paper.Thirdly,a systematic optimized model including the computing time optimized sub-model,energy consumption optimized sub-model,reliability optimized sub-model,security optimized sub-model and cost optimized sub-model was proposed to meet the corresponding requirement of the users’.Finally,a high-order matching model was analyzed,which is based on the compu-tation of HOSVD.The integration of different tensor spaces according to the same order is realized.Then,a high-order matching model is proposed to analyze the requirement,interests of the users in different conditions.Accordingly,the personalized service will be provided. |
PDF Full Text Request