【作者】 包欢；

【导师】 孙付平；

【作者基本信息】 解放军信息工程大学 ， 测绘科学与技术， 2009， 博士

【摘要】 随着社会经济的发展,我国建筑业得到迅猛发展,各种高、大、特、新的建筑物不断涌现。众所周知,建筑物从施工开始起,就会由于自然和人为的因素产生各种变形。了解变形状况,分析形变原因,预报未来变形,对于预防事故,保证建筑物正常使用是非常重要的。随着人们对建筑物形变监测重要性认识的不断深入,以及国家相关法律法规的实施,建筑物的形变监测已越来越广泛地应用于实际工程中,开发大型建筑物形变监测系统也越发显得重要。本文针对大型建筑物形变监测中的热点和难点问题进行了研究,主要研究内容和创新点如下:1.介绍了建筑物形变监测系统的研究意义和现状,研究了监测系统的总体方案设计、系统的软硬件平台以及系统的通讯及远程控制方法,并阐述了TPS数据采集器关键部件与其功能,对系统的通讯及控制系统也进行了详细的分类,并指出了各自的优缺点与适用环境。2.分析了测量误差的来源、分类和处理方法,综合讨论了对监测数据进行预处理以保证数据可靠性的意义。3.对目前粗差处理的理论和方法进行了系统的分类总结,对常用的几种粗差探测方法进行分析比较,得出一些有益的结论。并对关联分析法探测粗差提出了一种简单的多项式模型,实例表明,该方法简单易行,检验结果的可靠性高。4.应用过程突变理论建立了监测数据的动态检验模型,具有计算量小,速度快的特点,在监测点较多的监测项目中十分有效。在实际应用中,结合关联分析检验法,可以有效地解决自动化监测系统对监测数据真实性的识别问题,从而提高了自动化监测系统的实用性。5.详细讨论了人工神经网络的基本原理和建立监控模型的具体方法,对BP神经网络存在的问题进行了分析讨论,并引入了采用非线性规则化函数对原始数据序列进行预处理以及记忆初始权值、阈值的方法,大大提高了BP模型的收敛速度,并防止了模型陷入局部极小值,提高了模型拟合的精度。6.分析了系统误差产生的原因,介绍了系统误差的常用的几种检验方法,指出了各种检验方法的优点与其不适用性。7.针对极坐标测量中存在的系统误差,提出了多重实时差分的测量方案,对该方法的精度进行了理论分析,推导了3维坐标的中误差公式。分析了球气差系数的内涵,表明其主要是由大气垂直折光系数决定的。8.介绍了灰色系统理论中的GM(1,N)灰关联模型,并提出了应用灰色关联度来选择显著自变量,以提高模型拟合度的改进方法,并给出了改进的灰关联模型的解算流程。9.研究并建立了各观测方向折光系数的灰关联模型,根据灰关联模型快速求解所观测方向的折光系数,对观测的高差进行实时改正,在较大程度上有效地减弱了折光的影响。据此,首次提出了改进的极坐标实时差分测量方案,实际试验结果表明,该方案对大气垂直折光的改正效果较好,实际使用也较为方便,有效地提高了单向三角高程的测量精度。10.对经典卡尔曼滤波与扩展卡尔曼滤波(EKF)的算法模型与求解步骤进行了分析研究,讨论了各自的优缺点与适用范围。11.在建筑物形变监测领域首次引入了Sigma点卡尔曼滤波(SPKF)方法。在介绍Sigma点变换算法及分析其精度的基础上,详细推导了SPKF算法流程,并且通过实例说明了SPKF较之EKF方法的优点:不需要计算雅可比矩阵,提高了计算的效率,预报精度较高,性能更为稳定,不易造成滤波发散。12.首次提出了神经网络SPKF方法,并将其应用于建筑物形变监控这一非线性系统的建模预报中。这种方法解决了BP神经网络训练时间长,对初值依赖大、容易陷入局部极值的缺限,也将神经网络具有逼近任意连续函数和非线性映射能力的优点通过SPKF得以顺利实现。算例结果证明了这一思路和方法的可行性和有效性。13.通过作者近几年完成的几个有代表性监测项目的介绍,说明了形变监测系统理论在实际中的应用情况。首次全面分析了监测房玻璃折射对观测方向及观测距离的影响,对监测方案的设计具有指导意义。对于建筑物规模较大,监测区域较广的情况,首次设计了多测站实时形变监测系统,开发了相应的软件与数据库系统。14.对于特殊监测条件比如地铁,组建了多测站组成的监测网络系统,并提出了实时动态基准对向导线测量的创新方法,有效地解决了监测中的车挡目标、大气湍流、地基振动、多目标干扰等问题,填补了国内地铁结构变形监测领域的空白。通过实际应用表明,监测精度达到了项目要求。更多还原

【Abstract】 With the development of social economy, the building industry is growing significantly and as a result, high, large, unique and novel buildings emerge like mushrooms. As is well known, buildings undergo deformations due to natural and man induced causes from the start of ground-breaking, and therefore, it is very important to find out deformation status, analyze causations, and predict deformation trends for preventing incidents and ensure the normal use. As there is a deeper knowledge of the importance of deformation monitoring, and with the implementation of relative laws and regulations, the deformation monitoring of buildings has been applied in engineering practices more and more widely, which makes the development of deformation monitoring system for large buildings more and more important. The dissertation is aimed at some hot and difficult issues in the deformation monitoring of large buildings, and the main work is as follows:1. The significance and status of the research of deformation monitoring of buildings were introduced. Study was made on the integral structure of monitoring system, software and hardware of such systems, the design of system communication as well as remote control methods. Special attention was given to the key part as well as the function of TPS data collector. The system communication and control system were illustrated in detail and their advantages as well as shortcomings and applicable environment were discussed.2. The origins, categorization and processing methods of surveying errors were analyzed. The sense of pre-processing of monitoring data to ensure data reliability was discussed from a general viewpoint.3. Systematic categorization and summing-up was made for current theory and methods of outliers processing. Analysis and comparisons were made on some commonly used outlier detection and some beneficial conclusions were drawn. Especially, a simple polynomial model was proposed for outlier detection based on correlation analysis method, and example shows that the method is easy to use and reliable in detection.4. Process Mutation Theory was applied for establishing the dynamic verification model of monitoring data, which is characterized by low computation workload and fast calculation. The model was proved effective in monitoring projects that involves a lot of points. In applications, the model combined with correlation analysis verification method can perform the identification of automated monitoring system for the authenticity of monitoring data and therefore improves the applicability of automated monitoring system. 5. Specific focus is put on the principle of Artificial Neural Network (ANN) and the method used in constructing monitoring model. For the problems with BP Neural Network, analysis was made and solutions were put forward to solve the problems. The dissertation proposed the method that applies nonlinear regularization function in the preprocessing of original data series and memorizing initial weight and valve values, which greatly improves the converging rate of BP model and prevents the model from reaching local minimum and thus raises the fitting accuracy of the model.6. The reasons why systematic errors are generated were analyzed. Several inspection methods of systematic errors were introduced. For the methods, their advantages and inapplicability situations were pointed out.7. For the systematic error hiding in polar coordinates measurement, a surveying scheme based on multi real time differencing method was presented, and theoretical analysis was made on the accuracy of the method, and furthermore, the Mean Square Error formulae in 3-dimensional coordinates were derived. The coefficients of the effect of Earth curvature and refraction were discussed in detail, which was proved to be chiefly determined by the coefficient of the vertical refraction of atmosphere.8. The GM(1,N) gray correlation model of the Gray System Theory was introduced. A method was proposed that improve model fitting using the degree of gray correlation in the selection of the independent variable of prominence, and a workflow of the solution of the improved gray correlation model was put forward.9. The gray correlation models for the refraction coefficients of observed directions were studied and set up, according to which the refraction coefficient for the observed direction can be solved readily, and real-time corrections can be made to the observed height difference, and thus the effects of refraction can be reduced effectively to a great extent. Therefore, the scheme of improved polar coordinates real-time surveying was presented for the first time. Field experiments demonstrated that the scheme can result in a better correction to errors of vertical refraction of atmosphere, and the application of the scheme was proved to be convenient. The scheme can effectively raise the precision of one-way trigonometric height surveying.10. Models of classic Kalman Filtering and the Extended Kalman Filter(EKF) as well as their solutions were introduced. The advantages and disadvantages and applicability of the two methods were also discussed in detail.11. The Sigma Point Kalman Filter(SPKF) method was introduced. Firstly the Sigma Point Transform was introduced and its accuracy was analyzed. Then the SPKF formulae were derived in detail. Examples showed the superiority of SPKF to EKF which are no need for the computation of Jacobi matrix, higher accuracy of prediction, more stable performance of filtering.12. Neural Network(NN) based SPKF was proposed for the first time and was applied in the modeling and prediction of nonlinear system of the deformation monitoring of buildings. The commonly used Back Propagation(BP) NN has some defects, like long training time, relying heavily on initial values and the proneness of reaching local extreme value although the NN has the basic advantage of approximating any continuous function and nonlinear mapping with high accuracy. However, the advantage of NN is embodied through the combination with SPKF. Experiment proved the applicability and effectiveness of NN based SPKF.13. Some representative monitoring projects that the author joined in recent years were introduced. A comprehensive analysis was made on the effect of the glass refraction of monitoring house which accommodates Total Station on the observed directions and observed distances, which can be used as reference for future monitoring projects. For large-scale buildings or extensive monitoring areas, the multi station real-time deformation monitoring system was constructed for the first time, and corresponding software and database systems were developed.14. For special monitoring situations like subway, the author joined in the construction of monitoring network system composed of multi stations. The creative method of real-time dynamic reciprocal traverse measurement was introduced, which filled the blank of domestic subway deformation monitoring field by solving the problems encountered in monitoring, such as blocked target, atmosphere torrent, ground base vibration, multi target interfering, etc. Filed applications demonstrated that the monitoring accuracy satisfied the demands of the project.更多还原