基于RBF核ε-SVR的导水裂隙带高度预测模型研究  

Predictable testing and determination of the height of the fractured waterconducting zone based on theε-SVR model via the RBF kernel function

作　　者：徐树媛[1] 张永波[1] 孙灏东 胡晓兵 XU Shu-yuan;ZHANG Yong-bo;SUN Hao-dong;HU Xiao-bing(School of Water Resources Science and Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Shanxi Coal Geological Bereau,Taiyuan 030006,China;Shanxi Coal Geology 114 Prospecting Institute,Changzhi 046011,Shanxi,China)

机构地区：[1]太原理工大学水利科学与工程学院,太原030024 [2]山西省煤炭地质局,太原030006 [3]山西省煤炭地质114勘查院,山西长治046011

出　　处：《安全与环境学报》2021年第5期2022-2029,共8页Journal of Safety and Environment

基　　金：国家自然科学基金项目(41572221);科技部国家重点研发计划项目(2018YFC0406403);。

摘　　要：为了构建更多样本条件下的高精度导水裂隙带高度预测模型,利用灰色关联理论筛选出关联度较高的煤层采厚、顶板岩层强度及其组合特征、采深、工作面斜长、开采分层数及倾角等因素作为输入变量;应用支持向量机回归模型理论,建立了不同模型类型与不同核函数组合的4种支持向量回归模型;采用PSO算法对模型参数进行优化,并对4组模型预测结果进行分析比较。结果表明,基于RBF核ε-SVR预测模型在预测精度、拟合效果与训练效率等方面优于其他3组模型。通过对PSO算法优化的RBF核ε-SVR模型与传统经验公式的预测效果进行综合比较和讨论,可得支持向量回归模型在现代化开采条件下预测准确率更高,可为矿井安全开采与资源综合开发提供依据。The given paper has the maximum height of fractured water-conducting zone to be the research target.For the said research goal,we used the support vector regression theory,an empirical model,to the establishment of a predictive model.In larger samples,for better prediction of the height of fractured water-conducting zone,first,we have selected 6 factors as input variables and have ranked them by grey correlation theory.The factors highly relevant to the development of fractures includes the mining height,the strata strengths and combinations of coal seam roof,mining depth,working face length,mining layers and dip angle of the coal seam.Then,we have established 4 prediction models based on the support vector regression with different model types and kernel functions.At the same time,we have also managed to optimize the parameters in accordance with the PSO method to confirm the forecast results for the height of the said zone.What is more,we have analyzed and compared the predicting results of each SVR model.The results of the predicted height in accordance with theε-SVR prediction model with RBF kernel function should be said superior to those gained from the other 3 models in terms of accuracy,fitting effect and training efficiency.Thus,the result helps us to determineε-SVR and the radial basis function(RBF)as the type and kernel function in the given support vector regression forecast model.In addition,the comprehensive comparison results help us to determine the prediction effect and its applicability on the height of the fractured water-conducting zone between the empirical equations and theε-SVR prediction model with the RBF kernel function.And,all of the above research has made us conclude that the empirical formula is unique available to traditional mining method,while the supportive vector regression model can help to lead to the higher prediction accuracy and broader application in the modern mining practice.

关 键 词：安全工程 支持向量机 回归 粒子群算法 导水裂隙带 径向基函数 预测模型 

分 类 号：X932[环境科学与工程—安全科学]