基于智能算法的滑坡位移预测与危险性评价研究

发布时间：2018-04-12 16:43

本文选题：智能算法 + 滑坡灾害　；参考：《中国矿业大学(北京)》2016年博士论文

【摘要】：滑坡灾害是一种常见的地质灾害,它的发生,不仅给人类的生命、财产安全带来严重威胁,同时给资源、环境、生态等各个方面带来了巨大破坏。我国是世界上滑坡灾害最为严重的国家之一,据中国地质环境信息网发布的中国地质灾害通报显示:2006至2015年10年间全国共发生地质灾害260353起,其中滑坡发生的比例居于地质灾害首位,占地质灾害总数的73.79%,这10年期间由滑坡造成的伤亡、失踪人数共计11281人,直接经济损失435.05亿元,不仅如此,滑坡所带来的次生灾害也是难以估计的。因此,采取必要的手段对其监测,进而科学、有效地对滑坡灾害进行预测预报,具有重大的经济价值及社会意义。然而,由于滑坡发展演化的影响因素众多(如地形地貌、地质构造、地层岩性、水文条件、降雨及人类工程活动等),使得滑坡运动具有开放性、复杂性和不确定性等特点,导致很难采用传统方法对其进行准确预测预报,为此,本文提出以“智能算法”为主要研究手段,并融入现代测绘数据处理、灰色预测、灰色决策等理论知识,以我国境内典型滑坡工程:链子崖滑坡、卧龙寺滑坡、古树屋滑坡、新滩滑坡和重庆万州区滑坡为研究对象,围绕“滑坡变形位移预测”和“滑坡危险性等级区划”,构建了基于智能算法的滑坡灾害预测预报体系,主要研究内容与成果如下:(1)滑坡变形位移状态辨识与曲线特征分类(1)自然界中赋存的滑坡体,由于其形成条件、诱发因素的不同,导致滑坡变形累积位移曲线形态各异,深入研究了滑坡变形累积位移曲线的特征,并按形态将其分为:减速-匀速型、匀速-增速型、减速-匀速-增速型、复合型四类;通过对链子崖、卧龙寺、古树屋、新滩四个典型滑坡体的地形地貌、地质构造、影响因素等分析基础上,对四个滑坡体的变形曲线特征进行辨识归类;(2)掌握滑坡变形位移曲线特征,对判断滑坡的成因模式、变形发展阶段、诱发因素影响程度,以及预测预报模型的选取,具有重要的指导作用。(2)基于经典智能算法BP、RBF的滑坡变形位移预测(1)针对标准BP算法网络收敛速度慢、易于陷入局部最小等缺陷,推导出了四类BP改进算法,以链子崖、卧龙寺、古树屋、新滩四类滑坡体为例,建立了基于BP改进算法的滑坡变形位移预测模型,深入讨论了BP算法建模时应注意的若干问题,给出了BP网络结构参数优化实施的具体流程,构建了BP滑坡变形位移预测的最优网络拓扑结构,算例表明,四种改进算法在预测效果方面较标准BP算法有明显改善,且LM-BP算法预测效果最优;(2)阐述了RBF神经网络的结构、训练算法及建模过程,提出了基于二维区间搜索的rbf网络参数优化方法;从隐含层传递函数、隐含层节点数量、训练算法、逼近方式等方面将rbf与lm-bp进行比较,以链子崖、卧龙寺、古树屋、新滩四类滑坡体为例,分析比较rbf、lm-bp算法用于滑坡变形位移预测的适用性,实验结果表明,rbf较bp算法,在网络收敛速度、网络泛化能力、外推预测方面有所改善。(3)基于新型智能算法elm的滑坡变形位移预测(1)将elm算法引入到滑坡变形预测中,深入剖析了elm算法的学习机理,指出了其与bp、rbf算法存在的本质区别,elm克服了经典算法bp、rbf采用梯度下降训练网络,导致网络易局部最小化的缺点;(2)基于误差处理视角,对elm网络输出权重参数β?的求解进行了推导,发现其求解过程是基于最小二乘估计,从而导致elm算法抵御粗差性能较差;为增强elm算法抵御粗差的能力,将广义极大似然估计(m估计)与之相融合,提出了基于m估计的robust-elm滑坡智能预测模型;(3)标准elm算法对滑坡数据中的粗差较为敏感,抗粗差性弱;基于m估计的robust-elm算法能较好的抵御滑坡数据中的单个、多个粗差,且预测精度较高。(4)基于智能耦合模型的滑坡变形位移预测针对采用单一智能算法进行滑坡变形预测时所存在的问题,提出构建滑坡耦合预测模型,基于三个视角,构建了三种形式的耦合模型:基于“权重约束”的耦合模型,基于“算法融合”的耦合模型,“顾及诱发因素影响”的耦合模型:(1)根据权重求取的不同约束准则,对“权重约束”耦合预测模型的构建进行研究,分别构建了“最优权”、“非最优权”、“灰色综合关联度定权”、“熵权”、“elm非线性权”五种形式的耦合预测模型,讨论了五类约束准则权重的求取特点,以古树屋滑坡、新滩滑坡为例,比较了五种约束准则下耦合预测的效果,算例表明,基于elm的非线性耦合预测具有良好的特性和较高的预测精度;(2)构建“算法融合”耦合预测模型时,首先,利用灰色累加生产算子,弱化滑坡变形位移数据的随机性,构建了基于灰化层处理的elm耦合预测模型;其次,顾及灰色模型群预测时所提供的有效信息,构建了基于灰色模型群的耦合预测模型,以古树屋滑坡、新滩滑坡为例,对两种形式的“算法融合”耦合预测模型进行了效果验证,其预测效果均较好;(3)以新滩滑坡、三峡库区某滑坡为例,建立了顾及降雨、库水位等诱发因素影响的多因子耦合预测模型。该模型首先将滑坡变形位移分解为趋势项和随机项,利用gm(l,l)模型提取变形的趋势项,然后采用elm算法逼近诱发因素与位移随机项间的非线性映射关系,算例表明,该耦合模型能结合滑坡监测数据的特点,从数据分解角度出发,兼顾了数据的趋势性与随机性,同时考虑了滑坡体的诱发因素,从多角度充分利用了滑坡监测数据的有效信息且预测精度较高。(5)基于多因素加权灰靶决策理论的ELM滑坡危险性评价以重庆万州区滑坡为例,首次将多因素加权灰靶决策模型引入到滑坡危险性评价中,并将其与新型智能ELM算法进行耦合,构建了顾及多影响因素加权灰靶ELM的滑坡危险性评价模型:(1)以滑坡灾害形成条件、诱发因素为出发点,选取高差、坡度、滑体物质类型、降水、人类工程活动等11项因素作为滑坡危险性评价指标,通过灰色关联分析得出影响因素权重的大小;(2)以灰靶决策理论为基础,根据所得靶心距的大小,对滑坡的危险等级进行量化评价,将滑坡危险等级划分为高度、较高、中度、低度和较低5级;进而采用ELM算法对待评价滑坡体的靶心距进行预测,根据所得靶心距,实现待评价滑坡体的危险等级区划;(3)基于加权灰靶决策ELM模型进行滑坡危险等级区划时,较为全面的考虑了不同影响因素所提供的有效信息,能综合考虑滑坡影响因素及权重的分配,通过靶心距的比较实现危险等级的量化,根据评价数据靶心距与标准靶心的比较,划分了滑坡危险性等级,实现了多因素定性与定量结合的预测,提高了滑坡危险预测的科学性与准确性。
[Abstract]:Landslide is a common geological disaster, it happened, not only to human life and property safety brought serious threat to resources, environment, ecology and other aspects brought great destruction. China is the world's landslide is one of the most serious countries, according to the Chinese geological environment information network released Chinese geological disaster Bulletin shows: 2006 to 10 years in 2015 a total of 260353 geological disasters, which happened in the proportion of landslide geological disasters, geological disasters accounted for 73.79% of the total, for the 10 year period by the landslide caused casualties, missing a total number of 11281 people, the direct economic loss of 43 billion 505 million yuan, not only that, it is difficult to estimate is secondary the disaster caused by the landslide. Therefore, to take the necessary measures for the monitoring and scientific prediction of landslide disaster effectively, has great economic value and social significance However, because many factors affect Landslide Evolution (such as topography, geological structure, lithology, hydrology, rainfall and human engineering activities, etc.), the landslide movement has openness, complexity and uncertainty, it is difficult using traditional methods to accurately predict, therefore, is proposed in this paper. In "intelligent algorithm" is the main research means, and integrate into the modern surveying data processing, grey forecasting, grey decision-making theory knowledge, to our country the typical landslide: Lianziya landslide, Wolong Temple Landslide, gushuwu landslide, Xintan Landslide and landslide in Wanzhou District of Chongqing as the research object, around the landslide deformation prediction "and the" division of landslide risk rating, constructed the forecasting system of landslide hazard prediction based on intelligent algorithm, the main research contents and results are as follows: (1) the deformation of the landslide displacement state identification and song Line feature classification (1) landslide occurrence in nature, because of its formation conditions, induced by different factors, lead to accumulative deformation shapes of displacement curve of landslide, in-depth study of the deformation characteristics of the landslide accumulative displacement curve, and according to the form which can be divided into: speed uniform, uniform growth - type, uniform deceleration - growth type, composite type four; through the Lianziya, Wolong temple, the ancient house, four typical Xintan Landslide topography, geological structure, based on the analysis of influencing factors, characteristics of deformation curve of four landslide identification classification; (2) Master landslide displacement curve. The causes of the landslide deformation mode, development stage, the influence degree of predisposing factors, and the selection of forecast model, has an important guiding role. (2) the classic intelligent algorithm based on BP, RBF landslide displacement prediction (1) according to the standard BP algorithm for network The slow convergence speed, easy to fall into local minimum defects, derived four kinds of improved BP algorithm in Lianziya, Wolong temple, old house, the new four beach landslide as an example, to establish a prediction model of deformation of landslide based on BP algorithm, discusses some problems should be paid attention to when modeling the BP algorithm the specific process, BP network optimization of structural parameters of the given optimal network topology construction deformation of BP landslide displacement prediction, numerical examples show that the four algorithms in prediction results than the standard BP algorithm is improved, and the LM-BP algorithm to predict the optimal effect; (2) describes the structure of RBF neural network, training algorithm and modeling process, puts forward RBF network parameter optimization method based on two-dimensional interval search; transfer function in the hidden layer, hidden layer node number, training algorithm, approximation methods aspects of RBF and LM-BP were compared in lianziya, Wolong temple, the ancient house, four kinds of Xintan landslide as an example, analysis and comparison of RBF, LM-BP algorithm for landslide displacement prediction of applicability, the experimental results show that RBF is in the BP algorithm, the convergence rate of network, the network generalization ability, extrapolation improves. (3) a new intelligent algorithm of elm landslide deformation displacement prediction based on (1) the elm algorithm is introduced to the landslide deformation prediction, in-depth analysis of the mechanism of elm learning algorithm, and points out the essential difference between BP, in the RBF algorithm, elm overcomes the classical algorithm BP, RBF uses gradient descent training network, resulting in network local minimum defects (2); error processing based on the perspective of the output weights of elm network parameter? Solution is derived, the solution process is based on the least squares estimation, which leads to poor performance of the elm algorithm against outliers; to enhance the ability to resist gross error elm algorithm, generalized The maximum likelihood estimation (M estimation) and the integration of the proposed intelligent prediction model of landslide robust-elm estimation based on M; (3) the standard elm algorithm is sensitive to outliers in the landslide data, outlier is weak; single robust-elm M estimation algorithm can resist landslide data better based on the multiple the gross error, and prediction accuracy. (4) predict problems existed when the landslide deformation prediction based on single intelligent algorithm intelligent displacement coupling model is proposed based on the landslide, landslide coupling prediction model, based on three perspectives, a coupling model is designed in three ways: Based on the "weight constraint" coupling model based on the fusion algorithm, the coupling model of "coupling model" for inducing factors: (1) according to the different weight constraint criterion, the research of "constructing weight constraint" coupling prediction model, respectively. The construction of "optimal", "non optimal weight", "grey comprehensive relationship right", "entropy", "prediction model elm coupled nonlinear power" five forms, the paper also discusses the characteristics of five kinds of constraints and criteria weights, to gushuwu landslide, Xintan landslide as an example, comparison five kinds of constraints under the criterion of coupling prediction results, numerical examples show that the prediction accuracy of nonlinear coupled elm has good characteristics and high based; (2) to construct fusion coupling prediction model algorithm, firstly, using the grey cumulative production operator, weakening the randomness of landslide deformation displacement data, construct the prediction model of elm coupling ashable layer based processing; secondly, the effective information provided for grey model forecast, prediction model is built based on the coupled grey model group, to gushuwu landslide, Xintan landslide as an example, the fusion algorithm for the two kinds of "" Coupling prediction model verified, the prediction results are better; (3) the Xintan Landslide in Three Gorges Reservoir area, landslide as an example, established for rainfall forecasting model of multi factor coupling effect of reservoir level induced factors. Firstly, the landslide deformation is decomposed into trend item and random item by GM (L, l) model to extract the trend of deformation, and then use the elm algorithm and the random displacement approximation factor between the nonlinear mapping relation. Examples show that the coupled model can be combined with the characteristics of the landslide monitoring data, from the point of data decomposition, taking into account the trend and random data, considering the induced the factors of the landslide, from many angles to make full use of the high effective information and the prediction precision of landslide monitoring data. (5) evaluation of ELM landslide risk factors weighted grey target theory based on Chongqing Wanzhou District landslide as an example, the first A multi factor weighted grey target decision model is introduced into the landslide hazard assessment, which is coupled with the new intelligent ELM algorithm, constructed the landslide hazard evaluation for multi factor weighted grey target model of ELM: (1) to the landslide disaster forming conditions, inducing factors as a starting point, select the elevation, slope the sliding body, material types, rainfall and human engineering activities as the 11 factors of landslide risk assessment index, by using gray correlation analysis the factors that affect the size of the weight; (2) the grey target decision theory, according to the target distance, the quantitative evaluation for risk grade of the landslide, the landslide risk rating divided into high, moderate, low height and low level of 5; then ELM algorithm is used to evaluate landslide - target distance forecast, according to the target distance, realize the risk zoning for the evaluation of landslide; (3) based on the The ELM model of grey target decision-making of landslide hazard zoning, effective information comprehensively considering different factors that can provide, considering influence of distribution factors and weights, quantified by a comparison of off target distance to the level of risk, according to the evaluation data from target compared with the standard target, the division of landslide hazard the level, prediction of a combination of qualitative and quantitative factors, improve the scientificity and accuracy of landslide hazard prediction.

【学位授予单位】：中国矿业大学(北京)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P642.22

【相似文献】