基于改进型PCA与SVR的煤矿瓦斯涌出量预测研究

发布时间：2018-06-09 01:51

本文选题：瓦斯 + 改进主元分析　；参考：《太原理工大学》2017年硕士论文

【摘要】：瓦斯灾害一直以来就是煤矿开采行业的主要灾害之一,不仅会给企业带来巨大的经济财产损失,还会对煤矿工作人员的生命安全造成威胁,因此做好瓦斯防治工作至关重要。瓦斯涌出量指在矿井建设以及生产过程中从煤层涌入到开采空间的瓦斯量,它是体现瓦斯在煤层中的赋存情况的一个重要参数。在煤层中进行煤炭开采时,瓦斯涌出通常是引起煤与瓦斯突出、瓦斯中毒和瓦斯爆炸等灾难事故的最主要原因,也是决定矿井通风的主要指标之一,是瓦斯防治中的重要一环。瓦斯涌出量能够比较综合的体现一座矿井中的瓦斯情况,其预测结果的正确与否,将直接影响矿井的各种技术经济指标,更直接关系到煤矿工人的人身安全。因此,制定合适有效的瓦斯涌出量预测方法对指导矿井安全生产作业具有十分重要的意义。实际生产中矿井瓦斯的涌出受到多种复杂因素的影响,在对这种有多维数据特征的样本进行分类或回归预测的学习时,通常会遇到学习模型过于复杂、训练时间过长以及最终预测结果不准确等问题。因此,需要对原始数据进行适当地处理,降低维度,去除噪声和冗余。同时由于支持向量机模型(SVM)具有许多良好的性质而在机器学习领域大放光彩,尤其是其在小样本处理上的优势以及泛化性能强等特点非常适合用于对瓦斯涌出量的预测。因此,本文提出了一种基于改进型主元分析算法(wPCA)与支持向量回归机(SVR)相结合的瓦斯涌出量的预测方法,最后经过实验取得了不错的效果,证明了该方法的有效性。本文主要涉及到下面一些工作。(1)简要总结瓦斯涌出量预测研究现状和各种涌出量预测方法,介绍分析煤矿瓦斯涌出机理和影响因素。(2)对瓦斯涌出量的主要影响因素进行分析,确定预测所需特征指标的选取原则,根据该原则选取合适的瓦斯涌出量预测指标。(3)探讨主元分析这种多元统计方法的基本原理及其算法步骤,并对其在数据处理中存在的不足之处进行针对性的加权改进,使得改进后的主元分析处理数据更为客观,降维能力更出色,有助于后期的学习训练。由于对原始数据进行了降维处理,在样本容量较大时可减少训练所需的时间。(4)介绍并深入研究支持向量机,在多次实验的基础上,选择合适的核函数及参数,将改进型PCA处理后的数据导入SVR进行学习训练,构建基于改进型PCA算法与SVR算法相结合的瓦斯涌出量预测模型。(5)对得到的预测模型用MATLAB进行实例仿真,同时用矿山统计法、BP神经网络、支持向量回归机等其他预测算法对相同的样本数据进行预测并相互对比。结果表明本文提出的基于改进型PCA算法与SVR算法相结合的瓦斯涌出量预测方法的平均相对预测误差为1.36%,均方误差为0.01,小于其它模型在相同样本下的预测误差,实验取得了良好的效果。(6)最后利用MATLAB GUI设计编写了瓦斯涌出量预测软件平台,实现可视化操作,人机界面友好,使得涌出量预测操作简单可行。
[Abstract]:Gas disaster has been one of the main disasters in the coal mining industry. It not only brings huge economic property losses to enterprises, but also threatens the life safety of coal mine workers. Therefore, it is very important to do gas prevention and control. Gas emission refers to the influx of coal from coal seam to mining in the process of mine construction and production. Gas quantity in space is an important parameter reflecting the occurrence of gas in coal seam. When coal mining is carried out in coal seam, gas emission is usually the main cause of the disaster accidents such as coal and gas outburst, gas poisoning and gas explosion. It is also one of the main indexes to determine the ventilation of the mine, and it is the weight of the gas prevention and control. The gas emission can reflect the gas situation in a mine in a more comprehensive way. Whether the prediction result is correct or not, it will directly affect the various technical and economic indexes of the mine, and it is more directly related to the personal safety of the coal miners. Therefore, a suitable and effective method of predicting the gas discharge is set up to guide the safety production operation of the mine. It is of great significance. In actual production, the gas emission of mine is affected by a variety of complex factors. In the study of the classification or regression prediction of this multi-dimensional data feature, the problems of too complex learning model, long training time and inaccurate prediction results are often encountered. The original data is properly processed to reduce the dimension and remove the noise and redundancy. At the same time, the support vector machine model (SVM) has many good properties in the field of machine learning, especially its advantages in the small sample processing and the strong generalization performance are very suitable for the prediction of gas emission. In this paper, a prediction method based on the combination of modified principal component analysis (wPCA) and support vector regression (SVR) is proposed. Finally, the effectiveness of the method is obtained through the experiment. This paper has proved the effectiveness of the method. This paper mainly involves the following work. (1) a brief summary of the current status of gas emission prediction research and the various types of gas emission prediction are briefly summarized. The mechanism of gas emission and influencing factors are introduced and analyzed. (2) the main influencing factors of gas emission are analyzed, the principle of selecting the characteristic indexes of the prediction is determined, and the suitable gas emission prediction index is selected according to this principle. (3) the basic principle of the multivariate statistical method of principal component analysis is discussed. The algorithm steps and the weighted improvement of its shortcomings in the data processing, making the improved PCA more objective, better ability to reduce dimension and help to the later learning training. (4) (4) introduction and in-depth study of support vector machines. On the basis of many experiments, the appropriate kernel functions and parameters are selected, and the data after the improved PCA processing are introduced into SVR for learning and training. The prediction model of gas emission based on the combination of improved PCA algorithm and SVR algorithm is constructed. (5) an example of MATLAB is used for the prediction model obtained. At the same time, the same sample data are predicted and compared with other prediction algorithms such as mine statistics, BP neural network, support vector regression machine and other prediction algorithms. The results show that the average relative prediction error of the gas emission prediction method based on improved PCA algorithm and SVR algorithm is 1.36%, and the mean square error is 0.01, Less than other models in the same sample prediction error, the experiment has achieved good results. (6) finally, using MATLAB GUI design and compilation of the gas emission prediction software platform, the realization of visual operation, human-computer interface friendly, making the emission prediction operation simple and feasible.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD712.5

【相似文献】