独立分量分析理论及其在变形监测数据处理与分析中的应用研究

发布时间：2019-01-09 13:34

【摘要】：变形监测涉及工程地质、结构力学、计算机科学等诸多学科的知识,它是一项跨学科的研究,并已发展为一门多学科交叉的边缘学科。主要包括两个方面的内容：首先是把握工程建筑物的稳定性,为安全运行诊断提供必要的信息,以便及时发现问题并采取措施；其次是科学上的意义,包括变形的机理,进行反馈设计以及建立有效的变形预告模型。在变形监测方法和技术进步的同时,获得的监测数据也越来越丰富,这些监测到的数据可以为变形体的状态提供更多的信息,但是另一方面,使得变形分析变得更加复杂。为了提高变形分析及预报的准确度,一方面需要对这些数据进行误差处理,提高观测值的精度；另一方面需要利用信息融合技术对监测数据进行分析与处理,以便更准确的进行工程结构健康的诊断以及灾害的预测预报,由此,信息的分解与融合就成了变形分析的一个主要任务。独立分量分析是从多元统计数据中寻找其内在因子或成分的一种方法。它是基于盲信号的分离而发展起来的,其突出的优势在于对于原始的信号不需要有太多的先验知识,且能更灵活更有效地的表征信号的本质结构,把独立分量分析用于变形监测信号处理中,在统计意义上更能反映工程建筑物变形的本质特性。本文把独立分量分析方法引入到变形监测数据处理与分析当中,通过仿真实验与大坝监测实测数据对独立分量分析方法在信号分离及信号降噪处理方面进行了研究,并在此基础上,进行多元回归分析,建立变形预测回归模型。本文的主要研究工作分两个部分：(一)基于独立分量分析的信号处理 1.基于独立分量分析的理论,在统计独立原则的基础上,通过分析多维观测数据间的高阶统计相关性,找出相互独立的隐含信息成分,进而得到独立分量。通过仿真实验对信号进行处理,实验结果表明,得到的独立分量与源信号非常相似,只是次序和幅值不确定。 2.通过独立分量分析对大坝监测实测数据的处理,来说明此方法在实际中的应用效果。大坝变形监测数据信号受到水位、温度、时效以及噪声的影响,通过设定一定数量的信号接收器,以满足源信号数量小于等于接收器数量的要求,从而实现独立分量分析在信号去噪中的应用。 3.独立分量分析方法并不能区分有用信号和噪声,所以经过此方法分离出的信号,要根据时域、频域及其他相关先验知识来加以区分。本文实例中分离出的信号,通过先验知识以及与分量的特性对比,再把信号从时域转到频域进行,可以把噪声与有用信号进行有效区分。 4.把独立分量分析去噪与小波去噪的效果进行比较。引入信噪比、均方差以及相关系数作为评价指标,通过仿真实验和实测数据的处理与分析,独立分量分析去噪的效果优于小波去噪,提取的独立分量精度更高,去噪鲁棒性更强。 (二)基于独立分量分析的多元线性回归分析 1.主分量回归已经在很多领域得到了应用,本文中,在模拟数据的基础上,利用最小二乘方法、主分量回归以及独立分量回归对假定的回归模型进行求解,通过结果知道,独立分量分析也可以应用在多元线性回归方法中,而且由此方法获得的独立分量可以对因变量进行更好的解释。 2.利用大坝的实测数据,运用独立分量分析方法对监测信号进行处理,在噪声信号被识别并剔除以后,利用剩下的因子去进行多元线性回归计算并得到回归模型,在对实测值和预测值进行比较的基础上对此模型的预测精度进行评价。
[Abstract]:Deformation monitoring involves the knowledge of many subjects such as engineering geology, structural mechanics, computer science and so on. It is an interdisciplinary study and has developed into a multi-disciplinary, cross-cutting edge discipline. The method mainly comprises two aspects: firstly, the stability of the engineering building is grasped, the necessary information is provided for the safe operation diagnosis so as to find the problems in time and take measures; secondly, the scientific significance, including the mechanism of deformation, and a feedback design and an effective deformation forecasting model are established. At the same time as the deformation monitoring method and the technical progress, the obtained monitoring data is more and more abundant, and the monitored data can provide more information for the state of the deformable body, but on the other hand, the deformation analysis becomes more complex. In order to improve the accuracy of deformation analysis and prediction, on the one hand, it is necessary to process the data and improve the precision of the observation value; on the other hand, it is necessary to use the information fusion technology to analyze and place the monitoring data In order to make more accurate diagnosis of structural health and forecast of disaster, the decomposition and fusion of information become one of the main functions of deformation analysis. The independent component analysis is to find out a factor or component in the multivariate statistical data. The method is developed based on the separation of blind signals, and has the advantages that too many prior knowledge is not required for the original signal, and the intrinsic structure of the signal can be more flexibly and effectively characterized, and the independent component analysis is used for deformation monitoring signals. In the treatment, the deformation of the engineering building can be more reflected in the statistical significance. In this paper, the independent component analysis method is introduced into the deformation monitoring data processing and analysis, and the independent component analysis method is studied by the simulation experiment and the dam monitoring data. Meta-regression analysis to set up a deformation pre-set Regression model is measured. The main research work in this paper is divided into two parts: (1) based on the independent score The signal processing of the quantity analysis is 1. Based on the theory of independent component analysis, on the basis of the statistical independent principle, the multi-dimensional observation data is analyzed. the high-order statistical correlation among the independent implicit information components is found out. The experiment results show that the obtained independent component is very similar to the source signal, and only the order and the amplitude value are not determined. 2. The process of monitoring the measured data of the dam by the independent component analysis will be described. The monitoring data signal of the dam deformation is affected by water level, temperature, aging and noise, and a certain number of signal receivers are set so as to meet the requirement of the number of the source signals to be less than or equal to the number of the receivers, so as to realize the independent component. 3. The independent component analysis method can not distinguish the useful signal and noise, so the signal separated by this method is based on time domain, frequency domain and frequency domain. The signal separated from the example is compared with the characteristic of the component, then the signal is transferred from the time domain to the frequency domain, and the signal can be transferred from the time domain to the frequency domain. The noise is effectively distinguished from the useful signal. 4. Separate the stand-alone components The effect of de-noising and small-wave de-noising is compared. The signal-to-noise ratio, the mean square deviation and the correlation coefficient are introduced as the evaluation index. the vertical component has higher precision and stronger denoising robustness. (2) A multi-element linear regression analysis based on independent component analysis. The main component regression has been applied in many fields. In this paper, on the basis of the simulation data, the main component is returned with the least two-by-by method. and the independent component analysis can be used in the multi-component linear regression method, and the method is obtained by the method, and the independent component analysis method is used for processing the monitoring signals, after the noise signals are identified and removed, the rest and the regression model is obtained, and the measured value and the predicted value are obtained.
【学位授予单位】：中南大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN911.7;O212.1

【参考文献】