基于支持向量机的黄土矿区最大下沉预计模型

发布时间：2019-03-14 13:19

【摘要】：煤炭开采引起地表沉陷变形,其中最大下沉值是衡量地表沉陷变形的关键指标之一。在开采沉陷中采深、采高、煤层倾角、覆岩硬度和工作面倾向和走向长度等因素对最大下沉值有重要影响。黄土矿区的地表最大下沉值还须顾及黄土层厚度与特性等的影响。这些因素与地表最大下沉之间存在复杂的非线性关系,已有的最大下沉预计函数模型无法准确地反映这种复杂的非线性特征,因而预计精度和适用范围均具有很大的局限性。支持向量机可通过核函数将样本空间维数映射到一个维数足够高的特征空间中,把复杂的非线性问题变成线性问题。支持向量机还有惩罚机制对粗差数据进行自动剔除,保证了模型的精度和可靠性,可为本文建立黄土矿区最大下沉预计模型提供有效手段。本文根据支持向量机原理及特点归纳出最大下沉预计模型的构建步骤,依据该步骤先采用交叉验证法确定了模型的惩罚参数和核函数参数;再用迭代方法筛选样本,取得最佳样本,用该样本进行模型训练,确定预计模型的支持向量个数及系数等参数,最终构建了支持向量机回归模型。再根据建模流程,基于MATLAB平台的脚本语言开发了可视化程序,实现了模型训练与应用过程的封装,以简洁、易懂的界面形式供用户操作。将支持向量机回归模型与已有的最大下沉预计模型进行了对比分析,发现支持向量机回归模型的精度和可靠性优于其它函数模型。为进一步揭示模型变量间的变化规律,分别取输入变量的不同值,研究它们与最大下沉值之间的关系。结果表明:最大下沉值对采高、岩层硬度和煤层倾角这三个输入变量的敏感性要高于其它变量。另外,对预计精度高的和预计偏差较大的样本进行了分析,发现预计偏差较大的样本其宽深比和土厚比分别都小于0.3和0.35。反之,预计精度高的样本的宽深比和土厚比至少有一个在0.4-0.5 以上。本文构建的支持向量机回归模型可用于黄土矿区地表最大下沉值的定量预计,具有一定的推广应用价值。
[Abstract]:Coal mining causes surface subsidence deformation, in which the maximum subsidence value is one of the key indexes to measure the surface subsidence deformation. The factors such as mining depth, mining height, coal seam inclination, overburden hardness, working face tendency and strike length have an important influence on the maximum subsidence value in the mining subsidence. The influence of the thickness and characteristics of loess soil layer should be taken into account in the maximum subsidence value of loess mining area. There is a complex nonlinear relationship between these factors and the maximum subsidence of the earth surface. The existing models of the maximum subsidence prediction function can not accurately reflect the complex nonlinear characteristics, so the prediction accuracy and the scope of application have great limitations. Support vector machines (SVM) can map the dimension of sample space to a feature space with high dimension by kernel function, and turn the complex nonlinear problem into a linear problem. Support vector machine (SVM) also has the penalty mechanism to eliminate the gross error data automatically, which ensures the accuracy and reliability of the model, and can provide an effective means for establishing the maximum subsidence prediction model of loess mining area in this paper. In this paper, according to the principle and characteristics of support vector machine, the construction steps of the maximum subsidence prediction model are summarized. According to this step, the penalty parameters and kernel function parameters of the model are determined by means of cross-validation. Then the optimal sample is selected by iterative method. The model is trained to determine the number of support vectors and the coefficients of the predicted model. Finally, the support vector machine regression model is constructed. According to the modeling flow, the visual program is developed based on the script language of MATLAB platform, which realizes the encapsulation of model training and application process, and provides users with simple and easy-to-understand interface form. The support vector machine regression model is compared with the existing maximum subsidence prediction model. It is found that the accuracy and reliability of the support vector machine regression model is better than other function models. In order to reveal the variation rule of the model variables, the relationship between the input variables and the maximum subsidence value is studied by taking the different values of the input variables respectively. The results show that the sensitivity of the maximum subsidence value to the three input variables, mining height, rock hardness and coal seam inclination, is higher than that of other variables. In addition, it is found that the ratio of width to depth and the ratio of soil thickness to depth are less than 0.3 and 0.35, respectively, for the samples with high prediction accuracy and large prediction deviation. On the contrary, the aspect-depth ratio and soil-thickness ratio of the samples with high accuracy are at least 0.4 脳 0.5. The support vector machine regression model constructed in this paper can be used for quantitative prediction of the maximum subsidence value of the loess mining area, and has a certain value of popularization and application.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD327

【参考文献】