基于SVM耦合遗传算法的回采工作面瓦斯涌出量预测  

Prediction for gas-gushing amount from the working face of stope based on the SVM coupling genetic algorithm

作　　者：董晓雷[1,2] 贾进章[1,2] 白洋[1,2] 樊程程[1,2] 

机构地区：[1]辽宁工程技术大学安全科学与工程学院,辽宁阜新123000 [2]矿山热动力灾害与防治教育部重点实验室,辽宁阜新123000

出　　处：《安全与环境学报》2016年第2期114-118,共5页Journal of Safety and Environment

基　　金：国家自然科学基金项目(51374121);辽宁省高等学校杰出青年学者成长计划基金项目(LJQ2011028)

摘　　要：为了对回采工作面瓦斯涌出量进行预测,提出将支持向量机(SVM)与遗传算法(GA)相耦合。利用GA寻找SVM最优的惩罚参数c和核函数参数g,并结合SVM训练速度快且具有良好泛化性能的特点,建立了基于SVM耦合遗传算法的回采工作面瓦斯涌出量预测模型。煤层深度、煤层厚度、煤层倾角、开采层原始瓦斯量、煤层间距、采高、临近层瓦斯含量、临近层厚度、层间岩性、工作面长度、推进速度、采出率、日产量对瓦斯涌出量的影响是复杂的、非线性的,因而将其作为预测的影响参数。将瓦斯涌出量作为目标参数。分别将影响参数和目标参数作为GA-SVM的输入值和输出值进行训练,训练后的预测输出和期望输出之间的误差绝对值作为GA的适应度函数值进行参数优化。结果表明,该预测模型预测的最大相对误差为5.878 2%,最小相对误差为0.923 0%,平均相对误差为2.180 9%,相比耦合前及其他预测模型有更强的泛化能力和更高的预测精度。The paper is aimed at introducing a prediction model for gas gushing-out amount of the stope face based on GA- SVM neural network. As is known,the so-called GA（ Genetic algorithm） can help obtain a global optimal solution while avoiding falling back to the local minimum value so as to find the optimal fine parameter C and the kernel function parameter g of the SVM network. The prediction reliability of the SVM system can be made better than BP and other neural network systems through quick-training and the effective performance of generalization. In one word,it is necessary to establish a prediction model for gas gushing amount of the working face based on GA- SVM. Thus,the new model has been endowed with the advantages both of SVM and GA. The main influential factors for the gas gushing-out amount of the working face include： the depth,the thickness and the dip angle of the coal layer,the primitive gas content of the mine,the spacing of the mining layer,the height of coal mining,the gas content of near coal layer,the thickness of near coal layer,rock properties of the coal seam,the length of working face,the advancing speed,the rate of production,the daily output of gas emission quantity. The influential factors of the mining layer on the gas gushing amount is complex and nonlinear by nature.Therefore,it is necessary to use them as the influential parameters of prediction,with the gas gushing amount being the target parameter. In addition,we have also to use the influential parameters and target parameters as the input and output value of the GA- SVM through training,and then,it will be possible to use the error absolute value of the prediction output and the output expected as the fitness function value of the GA to optimize the target parameters. The research results show that the maximum relative error can be 5. 878 2% in case the prediction model of SVM can be combined with GA,whereas the minimum relative error should be 0. 923 0%,with the average relative error being 2. 1809%. Thus,compared with the u

关 键 词：安全工程 支持向量机 瓦斯涌出量 遗传算法 适应度函数 样本数据 

分 类 号：X936[环境科学与工程—安全科学]