多点灰色变形分析与预报方法研究

发布时间：2018-05-29 09:35

本文选题：变形预警 + 多点分析　；参考：《西南交通大学》2017年硕士论文

【摘要】：变形监测主要是利用监测仪器来获取工程体的连续变化序列,进而通过综合评价与分析等技术对变形体的发展趋势以及安全状态进行评估。随着观测手段的不断更新,变形监测由传统的周期性人工测量方式转变为物联网模式下多传感器、卫星定位、摄影测量与遥感、移动通讯等技术相融合的现代化监测方式。由单一的平面高程位移监测系统转化为温度、气压、应力、位移等多角度综合监测系统。本文主要针对变形监测系统在运营前期测量数据量小、信息贫乏等条件下进行多点变形分析与预报研究。在研究过程中笔者深入探讨了多点预测的多项关键技术,主要提出了三种适用性不同的预测优化方法:(1)、顾及传感器数据起算误差的多变量灰色变形分析与预报模型;(2)串联式残差多点灰色模型;(3)多点多尺度并联式变形预测模型。主要工作如下:1、针对多点灰色模型五种建模方式进行了对比分析研究,结果表明五种建模方式均能较好对沉降序列数据进行模拟与预测,但动态建模方式预测精度高于静态建模方式。2、通过仿真实验说明了常用去噪模型在数据量较少情况下会将部分信息当作噪声进行过滤,即过度去噪。传感器监测系统在运营前期观测数据噪声会对模型参数解算过程造成影响。本文提出利用最小二乘解算常数项而多元整体最小二乘解算误差项的联合平差方法来抑制起算数据误差对参数解算带来的影响,从而达到无偏最优解以改进建模精度和预测精度。3、讨论了变形监测组合预警模型构建方式,将组合预测模型大致分为并联式组合和串联式组合两种预测方式。根据灰色模型建模机理构建了串联式残差多点灰色模型。深度解析了常见变形监测序列的表现形式,利用经验模态分解对变形因子序列与实际沉降序列进行提取,采用支持向量机与多点灰色模型分别对变形因子序列和实际沉降序列进行预测后重构得到最终预测序列。该组合模型弥补了多点灰色模型仅对近指数生长序列具有较好模拟效果的缺点。
[Abstract]:Deformation monitoring mainly uses the monitoring instrument to obtain the continuous change sequence of the engineering body, and then evaluates the development trend and the safety state of the deformable body by comprehensive evaluation and analysis. With the continuous renewal of observation means, deformation monitoring has changed from the traditional periodic manual measurement method to the modern monitoring method which combines multi-sensor, satellite positioning, photogrammetry and remote sensing, mobile communication and other technologies under the Internet of things mode. From the single plane elevation displacement monitoring system to the temperature, pressure, stress, displacement and other multi-angle comprehensive monitoring system. This paper mainly focuses on the multi-point deformation analysis and prediction of deformation monitoring system under the condition of small amount of measurement data and poor information in the early stage of operation. In the course of the research, the author deeply discusses several key techniques of multipoint prediction. In this paper, three kinds of prediction and optimization methods with different applicability are put forward, one is the multivariable grey deformation analysis and prediction model considering the starting error of sensor data, and the other is the series residual multi-point grey model, which is a multi-point and multi-scale parallel deformation prediction model. The main work is as follows: 1. The five modeling methods of multi-point grey model are compared and analyzed. The results show that the five modeling methods can well simulate and predict the settlement sequence data. But the prediction accuracy of dynamic modeling method is higher than that of static modeling mode. The simulation results show that some information is filtered as noise in the case of less data, that is, excessive denoising. The noise of sensor monitoring system will affect the calculation process of the model parameters. In this paper, a combined adjustment method of least square solution constant term and multivariate global least square solution error term is proposed to suppress the effect of starting data error on parameter solution. Thus the unbiased optimal solution is achieved to improve the modeling accuracy and prediction accuracy. The construction of deformation monitoring combined warning model is discussed. The combined prediction model is roughly divided into two kinds of prediction methods: parallel combination and series combination. According to the grey model modeling mechanism, the series residual multi-point grey model is constructed. The expression form of common deformation monitoring series is analyzed in depth, and the deformation factor series and the actual settlement series are extracted by empirical mode decomposition (EMD). Support vector machine (SVM) and multi-point grey model are used to predict the deformation factor series and the actual settlement series respectively. The combined model makes up for the shortcoming that the multi-point grey model only has a good simulation effect on the near-exponential growth sequence.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU196.1

【参考文献】