智能神经网络及其在隧道运营期变形预测评估中的应用

发布时间：2018-02-11 09:28

本文关键词： 智能神经网络变形预测 BP神经网络群组合模型隧道安全评估　出处：《西南交通大学》2015年硕士论文　论文类型：学位论文

【摘要】：变形监测包括对变形体的变形现象进行持续观测、对变形体变形性态的分析和对变形体的发展态势进行预测及安全评估等内容。随着信息化测绘的发展,变形监测观测手段由传统的人工测量向基于卫星定位、多传感器、互联网、移动通信等多项技术融合的物联网模式的现代化监测方式转变,如何在多源海量监测数据下对工程的安全状态进行预测评估是目前面临的最大挑战。特别是受经济条件及技术水平所限而修建的盐水沟隧道,其运营安全关系到整个西气东输重点工程的顺利进行,由于隧址环境恶劣、地质条件复杂并且已存在多处安全隐患,其安全严峻状况理应得到足够重视。然而,目前我国仍没有一部系统化的隧道安全评价行业规范,在实际中只能形成简单框架型的评价思路,可操作性十分有限。因此,科学地研究隧道变形预测与安全评估显得尤为重要。文章以新疆西气东输管道——盐水沟隧道变形监测工程为依托,以变形监测理论为基础,以数据分析为核心,借助人工神经网络与模型组合的优势,研究智能神经网络在运营期隧道变形预测分析与安全评估中的应用,对盐水沟隧道变形监测期间得到的大量监测数据进行了分析评估,保证隧道的运营安全。在论文研究过程中,笔者深入研究了各环节的关键技术,主要工作如下：1、针对标准BP算法的缺陷,研究动量-自适应学习速率算法、L-M算法及遗传算法对标准BP的改进,将三种改进方法与标准BP算法进行仿真实验对比分析,结果表明改进算法在收敛速度与拟合精度两方面均显著优于标准BP算法。2、将组合模型分成串联型、并联型与混联型三种组合方式,设计BP神经网络组合器,将三种改进模型与组合器以混联方式构建一种最优的智能神经网络模型—双重BP神经网络。实例分析表明,智能神经网络组合模型能有效提高单项模型的拟合精度。3、结合物联网模式下的多传感器融合隧道变形监测工程特点与难点,从基于时间序列和基于隧道变形影响因素两方面研究智能神经网络在隧道运营期变形预测中的应用,解决微空间灰色系统多关联因素下高精度隧道变形预测问题。4、研究隧道安全评价多层次指标体系构建及指标度量方法,设计五级评价集,用智能神经网络学习专家知识,对隧道运营安全评价指标进行逐级递归式的综合评判,根据评判结果确定运营隧道监测断面的安全等级。
[Abstract]:Deformation monitoring includes continuous observation of deformation phenomena of deformable bodies, analysis of deformability of deformable bodies, prediction and safety evaluation of deformability of deformable bodies, and so on. Deformation monitoring and observation means have changed from traditional manual measurement to modern monitoring mode of Internet of things based on satellite positioning, multi-sensor, Internet, mobile communication and so on. How to predict and evaluate the safety state of the project under the condition of multi-source massive monitoring data is the biggest challenge at present, especially the salt ditch tunnel, which is built due to the limitation of economic conditions and technical level. Its operation safety is related to the smooth progress of the whole key project of gas transmission from the west to the east. Due to the harsh environment of the tunnel site, the complex geological conditions and the existence of many potential safety risks, the serious security situation of the tunnel should be paid enough attention to. At present, there is still no systematic industry standard for tunnel safety evaluation in our country. In practice, it can only form a simple frame type evaluation thought, and its maneuverability is very limited. It is very important to study the tunnel deformation prediction and safety assessment scientifically. This paper bases on the deformation monitoring project of Xinjiang West to East Gas Pipeline-Salt Water Channel Tunnel, based on the deformation monitoring theory, and takes the data analysis as the core. With the advantage of combination of artificial neural network and model, the application of intelligent neural network in tunnel deformation prediction and safety assessment during operation period is studied. A large number of monitoring data obtained during monitoring of tunnel deformation in saline ditch are analyzed and evaluated. In the research process of this paper, the key technologies of each link are deeply studied. The main work is as follows: 1, aiming at the defects of the standard BP algorithm, This paper studies the improvement of the standard BP by the momentum adaptive learning rate algorithm and genetic algorithm. The three improved methods and the standard BP algorithm are compared with each other in the simulation experiment. The results show that the improved algorithm is superior to the standard BP algorithm in terms of convergence speed and fitting accuracy. The combined model is divided into three types: series type, parallel type and hybrid type, and BP neural network combiner is designed. An optimal intelligent neural network model, double BP neural network, is constructed by mixing the three improved models and combiners. Intelligent neural network combination model can effectively improve the fitting accuracy of single model. Combining with the characteristics and difficulties of multi-sensor fusion tunnel deformation monitoring engineering under the mode of Internet of things, intelligent neural network combination model can effectively improve the fitting accuracy of single model. The application of intelligent neural network in tunnel deformation prediction is studied based on time series and influence factors of tunnel deformation. To solve the problem of high-precision tunnel deformation prediction under multi-correlation factors in micro-space grey system, to study the multi-level index system construction and index measurement method of tunnel safety evaluation, to design a five-level evaluation set, and to use intelligent neural network to learn expert knowledge. The safety evaluation index of tunnel operation is evaluated by recursive method, and the safety grade of monitoring section is determined according to the result of evaluation.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U456.3;TP183

【参考文献】