【作者】 王妍；

【导师】 邓庆绪；

【作者基本信息】 东北大学 ， 计算机软件与理论， 2016， 博士

【摘要】 物联网是继计算机、互联网与移动通信之后的又一次信息技术革命与产业浪潮,它正成为经济社会绿色、智能、可持续发展的关键基础和重要引擎。以物联网融合创新为特征的新型网络化智能生产方式正塑造未来制造业的核心竞争力,工业、电力等行业应用仍然是物联网发展的重要领域。从近几年物联网的整体发展看来,智能信息处理依然薄弱。工业、电力等行业应用日趋复杂,这些领域的感知网络规模、基础设施等方面出现了新的特征,因此,物联网感知数据处理成为工业与物联网深度融合进程中的研究热点。本文深入分析和归纳了物联网新趋势下感知数据的特征以及感知数据处理面临的新问题,从物联网感知数据的获取、存储和查询等方面深入研究,取得了如下成果:(1)提出了一种频率自适应的数据采集方法。为了减少大规模物联网感知数据的采集量,降低数据传输消耗,本文基于感知节点密度对大规模感知网络进行划分,通过分析时间序列中采集数据的线性关系构建一元线性回归模型,根据采集数据的变化趋势,自适应的调整采集时间间隔,实验证明该方法显著地降低了数据采集量,减少网络能耗,具有较强的移植性。另外,该方法利用缺失数据估计模型填补缺失的数据,保证了感知数据获取的完整性。(2)提出了一种海量感知数据的泛在存储方法。为了实现物联网边缘智能化,提高数据处理实时性,降低网络传输负载,需要将部分海量感知数据存储于物联网的前端,因此本文提出物联网泛在存储模型和以分层扩展机制为核心的泛在存储方法。该机制采用扩展哈希编码方法动态地增加存储网元,避免突发、频发事件数据的丢失;采用多阈值级别方法将数据分散到多个存储网元上,避免了负载倾斜。实验证明该方法充分利用了物联网存储网元的存储资源,最大限度地满足海量感知数据的存储需求,实现了负载均衡的数据存储。(3)提出一种基于区域的大规模感知网络数据查询方法。目前,工业、电力等领域的物联网应用日益复杂,需要实时规划关键区域、重点关注区域或危险区域的监测,现有的物联网查询技术不能满足基于区域的灵活查询。因此,本文提出一种支持大规模感知网络任意区域查询的方法。该方法基于可变查询窗口实现大规模感知网络中任意区域的查询,利用映射数组代替实际查询窗口来减少查询下发时的通信消耗,通过建立临时查询树来解决查询结果的汇聚和转发。实验证明,该方法可以快速地查询并回收大规模感知网络任意局部区域的查询结果,不仅提高了查询实时性,并且能够大幅度地降低查询带来的数据传输能耗,延长感知网络寿命。(4)提出一种多维感知数据的高效Skyline查询方法。针对智能工业、智能电网等领域中多目标决策分析应用的实时性要求,本文提出一种多维感知数据的高效Skyline查询方法。该方法在查询下发时不携带过滤器,而是在收集查询结果时通过少量的计算逐步产生动态过滤元组,由此判断节点之间的支配关系,削减与查询无关的节点及其感知数据集,使网络中传输的非轮廓数据极少,且传输距离较短。该方法利用局部削减元组削减节点内的感知数据,进一步抑制非轮廓数据的上传。实验证明,该方法能够以较低的传输消耗,迅速返回被监测区域的轮廓数据,具有较好的可扩展性。(5)上述研究成果应用于辽宁排山楼金矿安全生产系统的工程实践。在该工程中建立了基于物联网分层结构的矿山安全生产系统,研制了基于频率自适应采集的监测监控系统,实现了感知数据的泛在存储、基于区域的实时查询和多目标决策分析。另外,针对矿山中人员定位系统定位精度不高等问题,提出了基于感知数据的动态目标追踪与预测方法。现有的动态目标定位技术多数仅适用于室内的二维空间,在工业物联网等领域存在电磁干扰和障碍物的复杂场景下,对动态目标的定位准确率较低,且无法预测位置。因此,本项目采用了一种基于感知数据的动态目标追踪预测方法。该方法根据感知数据建立预测模型,通过运动方程模拟动态目标运动特性,并利用卡尔曼滤波从估计值和观测值两个方面动态逼近真实值,预测下一时刻的位置。该方法在实际应用中不仅取得了较高的定位精度,实现了动态目标的位置预测以及动态目标进入危险区域时的预警,而且还能够分析出复杂施工环境下障碍物的分布状况。更多还原

【Abstract】 The Internet of things(IoT)is another information technology revolution and industry wave following the computer,Internet and mobile communication.It is becoming the key foundation and important engine of green,intelligent and sustainable development of economic society.The new networked and intelligent production,which is characterized by the integration and innovation of the Internet of things,is shaping the core competence of the future manufacturing industry.And these industry applications,mainly applied to industrial,electricity and other fields,are still pretty much part of the IoT’s development.From the perspective of overall development of the IoT in recent years,intelligent information processing is still weak,while industry,electricity and other industrial applications are increasingly more complex with many new characteristics that have been shown in the sensing network size,infrastructure and more aspects of these areas.So the sensing data processing of the IoT has become the research focus for the process of the highly integration of the industry and IoT.This paper analyzed and summarized the characteristics of sensing data and the problems faced by the sensing data processing under the new IoT trend.Taking much deeper research in the acquisition,storage and query of data located on Internet of things,the main contributions of this dissertation are as follows:(1)A data gathering strategy based on frequency-adaptive was proposed.To reduce the collection capacity and the transmission consumption of sensing data in the large-scale Internet of things,this dissertation first divides the large network based on node density;then establishes one-dimensional linear regression model by analyzing the linear relationship of collected data in the time series.According to the changing trend of collected data,the strategy adaptively adjusts the time intervals of data acquisition.Experiments show that this strategy significantly reduces the amount of collected data and energy consumption,and had strong portability.Also,this strategy uses missing data estimation model to fill in missing data to ensure data integrity.(2)A ubiquitous storage method of massively sensing data was proposed.To realize the intelligent at the edge of the IoT,improve the real-time performance of data processing and reduce the load of network transmissions,it needed to make part of mass sensing data stored in the front end of the Internet of things.Therefore,a ubiquitous storage model and a method with the hierarchical expansion mechanism as core were proposed.In this mechanism,the extended hash coding was adopted to dramatically increase storage network element to avoid sudden or frequent events data loss and the multi-threshold levels method was used to distribute data to multiple storage network element to avoid load skew.Experiments show that this method makes full use of the storage resource of storage network element in the IoT,maximally meets the storage requirements for the massive sensing data,and obtains better load balancing of data storage.(3)A data query algorithm based on region for the large-scale sensor network was proposed.At present,the application of Internet of things in industry and power is increasingly complex,which requires real-time planning of the monitoring for key areas,important areas which stress should be laid on,or dangerous areas.However,the current query methods of Internet of things cannot meet the flexible query for arbitrary region.Therefore,a query algorithm for variable region on large-scale sensor network was proposed.This algorithm can query an arbitrary region in the large network based on the variable query window,and communication consumption can be reduced by using mapping array instead of actual physical window when the queries are sent down.Building temporary tree was proposed to solve the problem of data aggregation and forwarding.The experimental results show that the algorithmcanquickly query and return the query results of any query region for large-scale sensing network.This not only improves the real-time performance,but also reduces the cost of network communication significantly and improves the lifetime of the sensor network.(4)An efficient Skyline query method for multi-dimensional sensing data was proposed.To meet the real-time requirement of multiple objectives decision-making in the industry applications,e.g.intelligent industry and smart grid,this dissertation proposed an efficient Skyline query method for multi-dimensional sensing data.Instead of filters were carried when the queries were sent down,dynamic filter tuples were gradually built with a few calculations when the query results were collected,so that determining the dominance relation between the nodes to filter out any node not relevant to the query and its sensing data sets.Thus,little non-spatial data was transmitted over the network and the transmission distance was shorter.Then the method used local-cutting tuples to filter out the sensing data inside the node to further control non-spatial data to be uploaded.Experimental results show that this method can quickly return the contour data of monitored area with lower transmission consumption and show good scalability.(5)The above mentioned theoretical results were applied to an engineering project of Paishanlou gold mine of Liaoning Province.Safety Production System in mine based on structure of IoT was established.Monitoring and controlling subsystem based on frequency-adaptive gathering model was developed.The Safety Production System achieved ubiquitous storage of massive sensing data,query based on region as well as multiple objectives decision-making.Since personnel position system has a poor positioning accuracy in mine,adynamic target tracking and predicting algorithm based on sensing data was proposed.Most of the existing dynamic target location technology is only suitable for indoor two dimensions.In the complex scene of electromagnetic interference and obstacles existed in the industrial Internet of things and other areas,they showed relatively low positioning accuracy of the dynamic target,and also cannot predict the location.So a dynamic target tracking and predicting algorithm based on sensing data was proposed.The algorithm established the forecast model based on sensing data.With simulating the motion characteristics of dynamic target by motion equation,Kalman filter was used to dynamical approach the true value from the estimated value and the observed value to predict the position of the target in the next moment.This algorithm has achieved higher location accuracy in the practical application,and realized the prediction of the position of the personnel and the early warning of dangerous areas.It can also successfully analyze the distribution of obstacles under complex construction environment.更多还原