因子分析法与BP神经网络耦合模型对回采工作面瓦斯涌出量预测

(1.西安科技大学 安全科学与工程学院,陕西 西安 710054; 2.西安科技大学 西部矿井开采及灾害防治教育部重点实验室,陕西 西安 710054)

瓦斯涌出量预测; 因子分析法; 因子选取; BP神经网络; 网络训练

Gas emission prediction in mining face by Factor Analysis and BP neural network coupling model
XU Gang1,2,WANG Lei1,JIN Hong-wei1,2,LIU Pei-dong1

(1.College of Safety Science and Engineering,Xi'an University of Science and Technology,Xi'an 710054,China; 2.Key Laboratory of Western Mine Exploitation and Hazard Prevention,Ministry of Education,Xi'an University of Science and Technology,Xi'an 710054,China)

gas emission prediction; Factor Analysis; factor selection; BP neural network; network training

备注

针对工作面瓦斯涌出量的影响因素众多且难以筛选的问题,提出了基于因子分析法与BP神经网络的工作面瓦斯涌出量预测方法。首先运用因子分析法对矿井瓦斯涌出量的影响因素降维处理,并筛选出3个主因子作为BP神经网络的输入端神经元,然后构建出基于BP神经网络的工作面瓦斯涌出量预测模型,并进行网络训练,最后对预测模型的可靠性进行检验。结果 表明:因子分析处理后变量作用在影响因子上的权重得到了重新分配,并且变量的维数得以减少,错综复杂的变量关系被优化成3个主因子之间的线性组合关系,使得BP神经网络模型预测的瓦斯涌出量结果更合理,精度更高; 工作面瓦斯涌出量预测值与实测值的相对误差均在5%以下,平均相对误差为3.25%,误差波动范围小,稳定性较好,为复杂因素影响下的工作面瓦斯涌出量预测提供了一条新的思路。

To solve the problem that the influence factors of gas emission in working face are numerous and difficult to screen,a prediction method of gas emission in facewas proposedbased on factor analysis method and BP neural network.Firstly,the factor analysis method was used to reduce the influence factors of mine gas emission with three main factors selected as the input neurons of BP neural network,and the prediction model of gas emission in working facewas constructedbased on BP Neural network with the network training carried out preformed.The reliability of the prediction model was finally tested.The results showed that the weight of the variable situated on the influence factor after factor analysis was redistributed,the dimension of the variable was reduced,and the complex variable relationship was optimized into alinear combination between the three main factors,which made the gas gushing result predicted by the BP neural network model more reasonable and the precision was higher.The relative error of the surface gas out prediction value and the measured value was below 5%,the average relative error was 3.25%,the error range was small,the stability was better,which provides a new way of thinking for the prediction of the surface gas gushing volume under the influence of complex factors.