【作者】 张耀平；

【导师】 曹平；

【作者基本信息】 中南大学 ， 岩土工程， 2010， 博士

【摘要】 采空区是采矿工程产生的最主要的灾害形式之一,当暴露面积增大到一定程度时,将可能会诱发灾难性的事故。研究采空区的稳定性及其岩层移动冒落特征和规律意义重大。本文以作者负责的项目“龙桥铁矿采空区监测与项板岩石冒落规律研究”为研究背景,采用室内试验、数值模拟、理论分析、现场监测与调查相结合的方法,借鉴统计学、多源信息融合技术的思想,研究揭示龙桥铁矿矿体回采产生的开采效应及关键块体特征参数。系统地对现有的岩石力学参数取值方法存在的缺陷进行分析,给出了用稳健回归方法确定岩石抗剪强度的模型,并对各种岩体参数取值方法进行评述。针对岩石抗剪强度确定中图解法和最小二乘法(LS)存在的问题：(1)只适用于实验数据相关性较高的情况,不适用于数据离散并有异常值存在的情况；(2)在对试验法获得的试验数据进行处理时,传统的最小二乘法由于采用残差平方和,容易夸大试验数据中异常值的影响。鉴于此,提出了岩石抗剪强度参数的稳健(Robust)回归分析方法。该方法在实验数据相关性差、数据离散并有异常值存在的情况下,具备削弱数据离散和对异常值定位的能力,提高了估计参数的稳健性和可靠性。该方法以残差的绝对值之和代替残差平方和,并通过复形法求得力学参数,避免了异常值的二次项,可以有效地减少异常值的影响。研究表明,在试验数据的相关性较好时,两种方法的计算结果相差不大,但当试验数据的相关性较差,存在异常值时,稳健回归方法的计算结果要优于最小二乘法。另外,为方便及提高数值计算精度,测量了龙桥铁矿的地应力。在前期现场调查、现场原岩应力测定、室内力学试验及岩体力学参数工程处理的研究基础上,采用FLAC3D方法对龙桥铁矿空区形成过程及采空区稳定性进行模拟计算和预测分析,分析了空区围岩的主应力、正应力、剪应力及变形情况,初步得到了围岩体变形和失稳的规律。采空区稳定性数值模拟结果表明,采空区形成后,空区四周各角隅处将首先达到极限剪切破坏状态。随着采空区的增大,角隅处破坏区域逐步延深扩大,顶板中央拉应力分布逐渐明显,最终可能变为拉应力破坏。根据研究结论,指出该矿山后续开采设计和管理中应注意的问题,并提出了合理的开采设计建议。传统的有限元分析方法很难分析岩体大变形及块体的跨落情况,系统地将关键块体理论、极限平衡分析与可靠度理论相结合,研究关键块体跨落的可能性。在常规可靠度分析理论一次二阶矩方法和二次二阶矩方法中,需要对基本随机变量进行正态变换或当量正态化,必须已知随机变量的概率分布,有时比较麻烦,而且这种变换也是导致功能函数非线性从而成为误差的一种来源。为消除这些误差,本文引入一次渐近积分方法与二次次渐近积分方法,很好的解决了该问题,不再需将变量正态化,提高了计算精度。科学准确的理解关键块体的特征参数,如关键块体的位置、体积、滑动方向、下滑力及抗滑力,可以有效对采矿方案进行优化,最大限度的提高开采效益,有效防止灾害事故的发生。建立空区上块状岩体的特征分析方法,该方法结合块体理论、可靠度理论及矢量分析可以确定关键块体的大小、移动方向、滑动模式、块体的可靠性及其失效概率等。通过本文建立的模型,根据模型的思路和数学公式编制了Matlab程序进行计算,计算结果显示,在15个开采区域,可动块体有270多个,可动块体中安全系数小于2的关键块体有48个,其中安全系数小于1.5的关键块体有33个,占68.73%。并给出了每个块体特征参数、可靠度及可靠概率。将安全系数与可靠度比较指出了安全系数在分析中存在的缺陷,由于岩体的复杂性和不确定性,很难准确确定岩石力学计算参数,然而,岩体力学参数的确定直接影响着计算结果,考虑参数的随机性,引入随机可靠度理论对其进行研究。研究发现,用安全系数分析关键块体的稳定性存在很大的误差,甚至有时候会得出完全错误和相反的结论。根据2009年龙桥铁矿回采后采空区的实际空间分布情况,结合整体稳定性分析对几何建模工作提出的要求,与龙桥铁矿技术人员多次讨论、分析,在确定本研究中矿体开采状况的时空对应关系后,对矿整个矿山开采显现的规律进行研究,主要从最大主应力、最小主应力、产生的垂直位移及地表变形等几个方面展开。将意大利产的Detector新型探地雷达设备应用到龙桥铁矿采空区覆盖层厚度探测中,并论述了探测原理及具体方法。该设备天线为IDS TR40MHz屏蔽天线,最大探测深度可以达到50米。探测结果表明,该矿山空区垫层厚度大于25 m,满足国家安全生产规程规定的无底柱分段崩落法空区垫层厚度必须超过2个分段高度。讨论了三种GM(1,1)模型与最小二乘支持向量机结合的方法：并联型、串联型和残差型灰色最小二乘支持向量机预测模型。给出了岩石覆盖层冒落及碴石厚度预测的灰色最小二乘支持向量机模型的建模思路及关键步骤,并应用该模型对覆盖层冒落高度及碴石厚度进行预测,研究表明,3种灰色最小二乘支持向量机的预测结果优于单一的GM(1,1)和单纯的LSSVM,各项指标均有所降低,体现了多信息融合的优势。残差型模型要优于并联型和串联型模型,表明用LSSVM来描述覆盖岩层及碴石厚度序列的随机项是正确可靠的。在串联型模型中,该组合预测结果要优于单个GM(1,1)模型预测,在实际应用GM(1,1)模型进行预测时,不能够事先知道哪个模型最优,因此使用该模型可以减小预测的盲目性,有一定的实用价值。充分显示了GLSSVM这种新的信息处理和预测方法的优越性,可以在工程实践中加以推广。更多还原

【Abstract】 The goaf is one of the most important forms of disasters induced by mining engineering. As mining progresses, when the exposed area of cavity increases to a certain extent, it will induce a catastrophic accident. In this paper, the horizontal projec "goaf Monitoring and the roof of goaf and falling law in Longqiao iron mine" as the research background around mining-induced rock movement and goaf falling effect cover the themes of work, combined with indoor test, numerical simulation, theoretical analysis, field monitoring and survey methods, the iron ore mining Longqiao effect produced by mining and key block feature parameter were revealed.Shortcomings existing of methods to get rock mechanics parameters was analyzed. The rock with a robust regression method to determine the shear strength model was given, and the various methods of rock mass parameters is reviewed and studied their in metal mining on application. Aiming at solve problems in determining the shear strength of rock in the graphic method and Least Squares (LS):(1) It only applies to a higher degree of experimental data related to the situation but not to discrete data and the existence of outliers; (2) When the test data obtained by the experimentation method are analyzed, the effect of anomalous values in test data increase markedly because the quadratic sum of residual error is adopted in calculation by the least square method. In view of this, a rock of stability of shear strength parameters (Robust) regression analysis is proposed. This method is related to the experimental data in the poor, dispersion and the existence of outlier cases, and can be weakened the value of discrete and abnormal positioning of the ability to improve the robustness and reliability of estimated parameters. In order to reduce the effect of anomalous values in test data, the sum of residual absolute value is used to replace the quadratic sum of residual error in the robust regress method, then the quadratic of anomalous value could be avoided, and the effect of anomalous value could be reduced availably. It shows that the results of the two methods are same when the relativity of the test data is good, but the results of the robust regress method are better than the least square method when there are have the anomalous values in test data. In addition, to facilitate and improve the numerical precision, Longqiao iron stress was measured.Based on the pre-site investigation, site of the original rock stress measurement, indoor rock mechanical parameters of the mechanical testing and handling of works, goaf forming process and the stability of simulation and prediction analysis of the goaf surrounding the principal stress, normal stress, shear stress and deformation were analyzed using FLAC3D. The law of rock deformation and instability iron was initially obtained.Simulation results of goaf stability show that the formation of goaf, the goaf around the corner will be the first to reach the ultimate state of shear failure. As the mined area increases, the destruction of the corner region of extended gradually expanded roof became clear the central tensile stress, tensile stress may eventually become damaged. According to the conclusions, follow-up exploration of the mine design and management should pay attention to the problems and proposed a reasonable design of the proposed mining.Traditional finite element method is difficult to analyze large rock deformation and block cross-off situation, according to key block theory, limit equilibrium analysis and reliability theory, across key block down the possibility was researched. In the conventional theory of first order second moment reliability analysis methods and second order second moment method, the need for the basic random variables are normal state of transformation or the equivalent, must be known to the probability distribution of random variable, sometimes too much trouble, and this kind of transformation is also the leading non-linear performance function to be a source of error. To eliminate these errors, we introduce an asymptotic integration method and the second time asymptotic integration method, a good solution to the problem.The establishment of the goaf on the characteristics of rock block analysis, the method combines block theory, reliability theory and vector analysis can determine the key block size, moving direction, sliding mbde, the reliability and effectiveness of block probability and so on.Through proposed model, based on the model of thinking and mathematical formulas to calculate the preparation of Matlab program, the results showed that in 15 mining areas, movable block with more than 270, the movable block in the critical safety factor is less than 2 Block 48, in which the key factor of safety less than 1.5,33 bulk, accounting for 68.73%. And the characteristic parameters of each block, the probability of reliability and reliability were given.Iron ore mining under the 2009 Longqiao Goaf after the actual spatial distribution, combined with the overall stability of the analysis of geometric modeling request, and iron and technical personnel Longqiao discussions, analysis, in determining the present study ore mining conditions corresponding relationship between time and space, the mining of the mineral appear to rule throughout the study, mainly from the maximum principal stress, minimum principal stress, the resulting vertical displacement and surface deformation, and several other fronts.The Detector will be in Italy a new application of ground penetrating radar equipment to the iron ore mined area Longqiao cover thickness detection, and discusses the principles and specific methods of detection. The equipments for the IDS TR40MHz antenna at the antenna, the maximum detection depth reach up to 50 m. Detected results show that the mine Air District cushion thickness is greater than 25 m, to meet national safety rules of order the natural caving pillar-free space zone cushion thickness must be more than two segmentation height.This article discusses the three types of GM (1,1) model and least squares support vector machines approach:parallel, series and residual-type gray squares support vector machine model, and iron by Longqiao examples show that such a new information processing and forecasting method is feasible and effective. Research shows falling rock and overburden thickness of ballast stone Grey forecasting model of least squares support vector machine modeling ideas and crucial step, and apply this model to cover caving height and thickness of ballast stone to predict the results show that,3 gray squares support vector machines is better than a single prediction GM (1,1) and pure LSSVM, the indexes had been lower, reflecting the advantages of multi-information fusion. Residual-based model is superior to parallel and series models, which cover with LSSVM to describe the thickness of rock and stone ballast sequence of random item are correct and reliable. In the series-type model, the combination is better than a single prediction GM (1,1) model prediction, in the practical application of GM (1,1) model forecasts can not know in advance which model the best, so use the model blindness can be reduced forecast, there are some practical value. Fully shows GLSSVM this new method of information processing and forecasting superiority can be extended in engineering practice.更多还原