时空流计算模型及其在边缘计算环境下的验证

发布时间：2018-01-08 05:20

本文关键词：时空流计算模型及其在边缘计算环境下的验证　出处：《浙江大学》2017年博士论文　论文类型：学位论文

【摘要】：近年来,移动传感器和无线网络的迅猛发展催生了大量带有地理空间信息的移动数据,这些移动数据具有持续到达、内容无限增长的流质特征,被称为时空流数据。由于时空流数据具有上述特点,传统的空间数据库的理论技术方法难完成实时计算,时空流数据的实时处理、分析、挖掘成为了国内外GIS的研究的热点和难点。时空流的计算速率表示单位时间内所处理流数据元组的数出量,是评价时空流处理方法优劣的重要标准。本文围绕如何提高时空流数据实时处理的计算速率问题,从模型表达、数据组织、查询优化等方面出发,构建基于基调代数的移动对象流质数据模型,设计应对频繁更新时空流概要数据结构,并提出利用边缘计算框架的流式空间大数据分布式空间计算方法,从而建立一套包含时空流数据模型表达、空间算法实现和时空流大数据并行化、边缘计算环境下时空流中心分散化处理的理论、技术、方法体系,为提高时空流计算速率提供一套完整的解决方案。本文的具体研究内容如下:(1)详细阐述基调代数表达方法,分析基调代数表达的移动对象抽象数据模型类型体系和操作算子,区分了离线数据和在线数据两种数据类型,提出从离线移动抽象数据类型向在线时空流数据类型的提升方法,构建时空流数据模型。(2)针对无约束空间和路网约束条件下移动对象运动特征,设计两种不同抽象数据模型离散化方法。为了达到提高时空流更新、查询计算速率的目的,无约束空间下,在传统空间索引的基础上添加二级索引完成自底向上更新方式;路网约束空间下,设计相应的路网节点、边界表并提出基于更新消息的CKNN算法,最终分别实现路网约束空间和无约束空间下的时空流数据概要数据结构设计。实验结果表明,无约束环境下,空间驱动型空间索引(网格索引)比数据驱动型空间索引(R树索引)具有较高的计算速率,更加适合移动对象的频繁更新;路网约束环境下,本文提出的UCKNN算法与IMA算法和CKNN算法相比具有更高的计算输出率。(3)针对移动对象单数据集内部空间查询(范围查询、邻近查询)和多数据集之间的空间计算(空间连接查询),提出两种时空流大数据分布式计算方案,探索影响时空流数据处理计算速率的多个因素并验证方案的可行性,以提高时空流数据计算速率。实验证明,应用四叉树作为二级索引,在不同的分区网格大小、移动对象数量以及不同比例下的更新、查询混合流都具有较好的计算速率。(4)针对时空流数据源地理分布广泛产生的大量网络宽带消耗问题,指出传统云中心处理模式在实时计算上的不足引入边缘计算模式,基于此将边缘集群部署于北京、杭州、郑州和湖州四地,从网络传输效率、边缘集群数量和本地、全局任务比例等多个角度进行时空流处理计算速率的探索性实验,进一步探究在边缘计算模式下时空流实时处理的计算速率变化规律。研究及结果表明,本文提出的时空流抽象数据模型能够表达通用的时空流数据对象表达、移动对象空间索引能应对移动对象高度频繁的数据更新、时空流分布式处理方法能在大数据环境下实时完成空间数据计算、边缘计算范式下的时空流实时处理方案能有效的解决地域分布广泛的数据源产生的时空流数据的快速处理问题,具有重要的科学和现实意义。
[Abstract]:In recent years, the rapid development of mobile and wireless sensor network has spawned a lot with the geographic spatial information mobile data, these mobile data has continued to reach the infinite growth characteristics of liquid content, known as spatial data stream. Due to the space-time data stream has the above characteristics, methods of traditional spatial database to complete the real-time calculation. The temporal and spatial flow of real-time processing, data analysis, data mining has become a hot and difficult research of GIS at home and abroad. The calculation flow rate time per unit of time the data tuple number output, is an important standard to evaluate the merits of the time stream processing method. This paper focuses on how to improve the computing speed problem of temporal and spatial flow of real-time data processing from the model, expression, data organization, query optimization of construction of mobile object data model based on liquid tone algebra, designed to deal with frequent more A new spatial stream synopsis data structure, and puts forward the calculation method of frame flow space of large distributed data space by edge, so as to establish a spatio-temporal data model includes flow algorithm and expression of spatial temporal and spatial flow of large data parallel computing environment, the edge of space-time flow center decentralized treatment theory, technology, method and system for. To improve the temporal and spatial flow to provide a complete set of solutions for computing rate. The main research contents of this paper are as follows: (1) detailed tone algebraic expression method, analysis of moving object algebra expression tone abstract data model and type system of operators, the distinction between the offline data and online data of two data types, put forward methods from the offline mobile abstract data type flow data types to the online space, build the spatio-temporal data stream models. (2) for unconstrained and constrained space network under the condition of moving objects The motion characteristic of two kinds of abstract data model discretization method. In order to improve the temporal and spatial flow update, query computation rate, unconstrained space, add two level index based on traditional spatial index on the bottom-up update mode; network space constraints, design the network node corresponding to the boundary table CKNN algorithm is proposed based on the update message, finally realized respectively the time-space road network constrained space and unconstrained space flow design data summary data structure. The experimental results show that the unconstrained environment, space driven spatial index (grid index) than the data driven spatial index (R index) to calculate the rate is higher that is more suitable for frequent updates of moving objects; network environment constraints, the proposed UCKNN algorithm compared with IMA algorithm and the CKNN algorithm has higher calculation output rate. (3) for moving objects A single data set internal spatial query (range query, nearest neighbor query) calculation and between multiple data sets of space (spatial join query), put forward two kinds of temporal and spatial flow of big data distributed computing scheme, explore the various factors affecting the flow rate of the spatiotemporal data processing and calculation and validate the feasibility of the project, in order to improve the spatial data flow calculation speed. Experiments show that the application of the four fork tree as the two level index, the partition grids of different sizes, the number of moving objects and update under different proportion, has better query rate calculation of mixed flow. (4) a large number of broadband network according to the geographical distribution of temporal and spatial flow of data sources from consumption problems, points out the deficiency of the traditional cloud processing center the model calculation in real-time using edge computing model, based on the edge of the cluster deployment in Beijing, Hangzhou, Zhengzhou and Huzhou four, the network transmission efficiency, the number of clusters and the edge Land, space and time stream processing experiments calculated by the rate of the angles of the global task scale, to further explore the calculation of rate changes in numerical model of temporal and spatial flow of real-time processing in the edge of the research and the results show that the proposed space-time flow model can be used to express the abstract data flow data object expression of general temporal, spatial indexing of moving objects to cope with the highly frequent updates of moving objects, temporal and spatial flow of distributed processing method in the environment of big data, real-time spatial data, temporal edge computing paradigm under the flow of real-time processing scheme can quickly deal with the problem of space and time to produce effectively solve the geographical distribution of a wide range of data sources of data streams, has important scientific and practical meaning.

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP311.13;P208

【相似文献】