基于SVM的煤与瓦斯突出危险性区域预测及防突技术研究

发布时间：2018-05-03 19:09

本文选题：煤与瓦斯突出 + 支持向量机　；参考：《中国计量学院》2013年硕士论文

【摘要】：煤与瓦斯突出是一个受多种因素综合影响的、复杂的非线性问题，用传统方法对其进行预测有很大缺陷。随着计算机技术和信息处理技术的快速发展，很多智能方法和技术也逐渐渗透到了类似于突出预测的一些问题中，，其中支持向量机是一种基于统计学习理论的机器学习方法，主要用于解决小样本、非线性、高维数、局部极小值等实际问题，并且具有良好的分类识别效果，已被广泛应用到众多领域的模式识别和预测预报中。为此，本文提出以现场和实验室检测数据为基础，通过引入支持向量机建立学习模型，实现突出危险性的分类预测。由于突出影响因素众多，不易区分突出发生的必要条件。因此，必须对原始数据进行预处理，以便获得影响突出的关键因素。为了有效解决该问题，本文采用灰色关联分析与熵权法结合的方法从原始样本中提取关键的特征指标。通过关键指标选取预测模型的训练和测试样本，并在此基础上建立支持向量机预测模型，其中整个模型的训练及测试过程在MATLAB平台下完成，并调用了LIBSVM软件包中的部分函数进行仿真程序的设计。另外，本文从支持向量机自身核函数选型以及参数优化的角度，对模型分类准确性的影响进行进一步研究，验证基于径向基（RBF）核函数更适合用于煤矿的突出分类预测。在此基础上分别通过交叉验证法和遗传算法对支持向量机的惩罚参数C和核参数g进行寻优，证明遗传算法能够在两个参数优选的前提下取得更好的测试效果。最后利用支持向量机的分类预测方法建立五阳煤矿南丰扩区76、78采区的区域危险性预测模型，测试结果与实际突出危险性情况相符。因此，该支持向量机模型可被用于采区未知区域的突出危险性预测。另外，本文结合“四位一体”综合防突措施，针对五阳煤矿提出以瓦斯预抽为主的防突措施。最后，通过超前钻孔进行防突措施有效性分析和检验。为实现该矿今后的煤与瓦斯突出综合防治提供方向。
[Abstract]:Coal and gas outburst is a complex nonlinear problem influenced by many factors. It has great defects to predict coal and gas outburst by traditional methods. With the rapid development of computer technology and information processing technology, many intelligent methods and techniques have gradually penetrated into some problems similar to prominent prediction, among which support vector machine is a machine learning method based on statistical learning theory. It is mainly used to solve the practical problems such as small sample, nonlinear, high dimension, local minimum, and has good classification and recognition effect. It has been widely used in pattern recognition and prediction in many fields. Therefore, based on the field and laboratory data, a learning model based on support vector machine (SVM) is proposed to realize the classification and prediction of outburst hazards. It is difficult to distinguish the necessary conditions of protruding because of its numerous influence factors. Therefore, the raw data must be preprocessed in order to obtain the key factors that have a prominent impact. In order to solve the problem effectively, this paper uses the method of combining grey correlation analysis and entropy weight method to extract the key feature index from the original sample. The training and test samples of the prediction model are selected through the key indicators, and the prediction model of support vector machine is established on this basis, in which the training and testing process of the model is completed under the MATLAB platform. Some functions in LIBSVM software package are called to design the simulation program. In addition, from the point of view of kernel function selection and parameter optimization of support vector machine, this paper further studies the effect of model classification accuracy, and verifies that RBF-based kernel function is more suitable for coal mine outburst classification prediction. On this basis, the penalty parameter C and kernel parameter g of support vector machine are optimized by cross-validation method and genetic algorithm, respectively. It is proved that the genetic algorithm can obtain better test results under the premise of optimal selection of two parameters. Finally, the regional hazard prediction model of 76YO78 mining area in Nanfeng expansion area of Wuyang coal mine is established by using the classification and prediction method of support vector machine. The test results are consistent with the actual outburst hazard situation. Therefore, the support vector machine model can be used to predict the outburst risk in unknown areas. In addition, combined with the comprehensive anti-outburst measures of "four in one", this paper puts forward some measures to prevent outburst in Wuyang coal mine. Finally, the effectiveness of anti-outburst measures is analyzed and tested through advance drilling. It provides the direction for the comprehensive prevention and control of coal and gas outburst in the future.
【学位授予单位】：中国计量学院
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TD713

【参考文献】