基于数据挖掘技术的瓦斯涌出量预测方法研究

发布时间：2018-06-19 08:20

本文选题：数据挖掘 + 支持向量回归机　；参考：《内蒙古科技大学》2013年硕士论文

【摘要】：瓦斯灾害是煤矿的主要灾害之一，不仅会给工作人员的生命安全造成很大的威胁而且还会带来大量的财产损失。做好瓦斯防治工作至关重要，而瓦斯涌出量预测是瓦斯防治中的重要环节，在很大程度上影响着煤矿的安全生产。因此，根据不同的矿井的情况，选择与其适应的瓦斯涌出量预测方法对指导矿井安全作业和制定行之有效的瓦斯灾害治理措施具有十分重要的意义。本文采用数据挖掘的相关技术与方法，对某矿瓦斯涌出影响因素及涌出量进行分析与预测研究，并利用Poly Analyst与MATLAB软件进行建模与计算。（1）运用Poly Analyst与MATLAB分别对瓦斯涌出量及影响因素进行相关分析与灰色关联分析，并得出以下结果：①某矿的瓦斯涌出量与煤层瓦斯含量、煤层埋深、煤层厚度、开采强度、邻近层瓦斯含量成正相关，与推进速度和工作面采出率成负相关；②以上各影响因素与瓦斯涌出量的相关系数的绝对值都在0.71以上，呈显著相关或高度相关。③经两种分析方法共同确定的影响因素为煤层瓦斯含量、煤层厚度、开采强度、煤层埋深、邻近层瓦斯含量。（2）利用Poly Analyst软件平台的支持向量回归机模型对瓦斯涌出量进行预测。通过对训练数据的预测，选取了支持向量回归机两种核函数的参数。由预测结果可知：①当多项式核函数参数及多项式的次数为5时，平均相对误差最小为0.91%。②当高斯核函数参数及标准偏差为2.1时，，平均相对误差最小为8.59%。（3）通过已选的两种核函数参数对测试数据进行预测，由预测结果可知：多项式核函数预测的平均相对误差为3.04%，高斯核函数的平均相对误差为5.39%，前者的预测精度优于后者。运用Poly Analyst支持向量回归机模型进行瓦斯涌出量预测，简单易行，便于掌握，能够充分运用瓦斯涌出量影响因素的各项数据，实现速度快、预测精度高，省去了大量繁琐的计算工作，并且能够取得良好的预测效果，为瓦斯涌出量的预测的实现提供了一条新途径。但是，在模型应用的过程中，要注意根据所研究对象的性质，选用合适的核函数及其参数。（4）利用MATLAB软件创建了一个满足网络设计要求的BP神经网络。通过对训练数据与测试数据的预测，可得出如下结论：①隐含层节点数的增加，虽可以提高网络的映射能力，但预测的精度不一定提高②对测试数据预测的精度较高，其最大相对误差为8.14%，平均相对误差为3.68%，误差小于10%满足精度要求。③为了使网络预测精度进一步提高，收集的样本数据应尽可能的多而准确，还要确定
[Abstract]:Gas disaster is one of the main disasters in coal mine, which will not only cause a great threat to the safety of workers, but also bring a lot of property losses. It is very important to do gas prevention and control work well, and the prediction of gas emission is an important link in gas prevention and control, which affects the safety production of coal mine to a great extent. Therefore, according to the situation of different mines, it is very important to choose the method of gas emission prediction to guide mine safety operation and to formulate effective measures for gas disaster control. In this paper, the related techniques and methods of data mining are used to analyze and predict the influencing factors and quantity of gas emission in a certain mine. Using Poly Analyst and MATLAB software to model and calculate. (1) Poly Analyst and MATLAB are used to analyze the quantity of gas emission and the influencing factors respectively. The following results are obtained: the amount of gas emission is positively correlated with the gas content of coal seam, the depth of coal seam, the thickness of coal seam, the intensity of mining, and the gas content of adjacent strata, but negatively with the speed of advancing and the mining rate of working face; (2) the absolute value of the correlation coefficient between the above factors and the amount of gas emission is above 0.71. The influencing factors determined by the two analysis methods are the gas content of coal seam, the thickness of coal seam, the intensity of mining. The gas emission is predicted by using the support vector regression model of Poly Analyst software platform. Based on the prediction of training data, the parameters of two kernel functions of support vector regression machine are selected. The prediction results show that when the parameters of polynomial kernel function and the degree of polynomial are 5, the average relative error is 0.91and 2.2. when the parameter and standard deviation of Gao Si kernel function is 2.1, The average relative error is the minimum of 8.59.) the test data is predicted by two selected kernel function parameters. The prediction results show that the average relative error of polynomial kernel function is 3.04 and the average relative error of Gao Si kernel function is 5.39.The former is better than the latter. The prediction of gas emission by using Poly Analyst support vector regression model is simple and easy to grasp. It can make full use of the data of influencing factors of gas emission and achieve fast speed and high prediction accuracy. It saves a lot of tedious calculation work, and can achieve good prediction results, which provides a new way to realize the prediction of gas emission. However, in the application of the model, we should pay attention to the selection of appropriate kernel function and its parameters according to the properties of the object studied. (4) A BP neural network that meets the requirements of network design is created by using MATLAB software. Through the prediction of training data and test data, we can draw the following conclusion: the increase of the number of nodes in the hidden layer of 1 can improve the mapping ability of the network, but the accuracy of prediction does not necessarily improve the accuracy of 2 pairs of test data prediction. The maximum relative error is 8.14, the average relative error is 3.68, the error is less than 10% to meet the precision requirement .3 in order to further improve the accuracy of network prediction, the sample data collected should be as much and accurate as possible, and must be determined.
【学位授予单位】：内蒙古科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TD712.5

【参考文献】