改进的灰色模型在煤矿地表沉降预测中的应用研究

发布时间：2018-11-20 20:16

【摘要】：随着我国的快速发展,对能源的需求与日俱增,促使煤矿的数量和规模不断扩大,此时,如何保证煤矿安全有效的开发和开采显得尤为重要,而煤矿地表沉降是其安全性的主要隐患,煤矿地表沉降的变形机理复杂、影响因数众多,并显非线性,如何准确有效地对煤矿地表沉降进行分析、监测和预报成为其首要任务。在地表沉降预测中灰色模型较为常见,而经典灰色预测模型中背景值的假定和初始值的选取存在不合理之处,并受原点误差影响不能做长期监测预报,针对以上不足之处,本文做了以下两点研究:1、本文在经典GM(1,1)预测模型的基础上,通过对其建模机理的深入研究,运用自动优选权重和最小二乘的方法,对背景值和初始值进行了改进。2、在对经典动态GM(1,1)预测理论进行研究的基础上,发现经典动态GM(1,1)预测存在维度选择的经验性和受原点误差性的不足,本文采用维度优选理论和实时引入监测值的方法对其进行优化,并建立了等维实时动态GM(1,1)预测模型。通过以某煤矿地表沉降监测数据为数据源,对以上改进模型进行了适应性研究。结果显示,以13#监测点为例,背景值和初始值优化后的GM(1,1)模型在拟合值精度上有大幅度提高,拟合值的残差在1mm到7mm之间,而用经典GM(1,1)模型得到的拟合值残差都在10mm之间;对预测值精度而言,经典GM(1,1)模型、背景值与初始条件优化的GM(1,1)模型、经典动态GM(1,1)模型、等维实时动态GM(1,1)模型预测值残差分别为15-48mm、11-39mm、11-12mm、1-3mm。测量方法的沉降误差为:M=±2.2(mm)。结果表明:1、背景值和初始值优化后的GM(1,1)模型在拟合值的精度上表现突出,预测值精度上也有一定的提高;2、等维实时动态GM(1,1)模型较适合做煤矿地表沉降预测;3、动态GM(1,1)模型较非动态GM(1,1)模型,在长期预测方面优势明显,更能反映实际情况。
[Abstract]:With the rapid development of our country, the demand for energy is increasing day by day, and the number and scale of coal mines are expanding constantly. At this time, how to ensure the safe and effective development and mining of coal mines is particularly important. The surface subsidence of coal mine is the main hidden danger of its safety. The deformation mechanism of the ground subsidence in coal mine is complex, the influence factors are numerous, and it is nonlinear. How to accurately and effectively analyze, monitor and forecast the surface subsidence of coal mine becomes its primary task. Grey model is more common in surface subsidence prediction, but the assumption of background value and the selection of initial value in classical grey prediction model are unreasonable, and can not be monitored and forecasted for a long time under the influence of origin error. This paper has done the following two studies: 1. On the basis of the classical GM (1Q1) prediction model, through the in-depth study of its modeling mechanism, the method of automatic optimal weight selection and least square method is used in this paper. The background value and initial value are improved. 2. On the basis of the study of the classical dynamic GM (1K1) prediction theory, it is found that the classical dynamic GM (1K1) prediction has the shortcomings of empirical selection and error of origin. In this paper, the theory of dimensionality optimal selection and the method of introducing monitoring data in real time are used to optimize the model, and a prediction model of equal dimension real-time dynamic GM (1 / 1) is established. Based on the ground subsidence monitoring data of a coal mine, the adaptability of the improved model is studied. The results show that the accuracy of fitting value of GM (1K1) model after optimization of background value and initial value is greatly improved, and the residual error of fitting value is between 1mm and 7mm, taking 13# monitoring point as an example. However, the residuals of the fitting values obtained from the classical GM model (1 ~ 1) are between 10mm; In terms of prediction accuracy, the residual values of the classical GM (1K1) model, the GM (1K1) model with optimized background and initial conditions, the classical dynamic GM (1K1) model, and the isodimensional real-time dynamic GM (1K1) model are 15-48mm / 11-39mm, respectively. 11-12mm 1-3 mm. The settlement error of the method is: M = 卤2. 2 (mm). The results show that: 1, the GM (1K1) model with optimized background value and initial value has outstanding accuracy of fitting value, and the precision of prediction value is also improved to a certain extent; (2) the iso-dimensional real-time dynamic GM (1t1) model is more suitable for ground subsidence prediction in coal mines, and the dynamic GM (1K1) model is more advantageous than the non-dynamic GM (1K1) model in long term prediction, which can reflect the actual situation more effectively.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TD327;N941.5

【相似文献】