基于经验模态分解和加权最小二乘支持向量机的采空区地面塌陷预测  

Ground Collapse Prediction of Mined-Out Area Based on EMD and WLS-SVM

作　　者：佴磊[1] 彭文[1] 袁明哲[2] 周能娟[1] 

机构地区：[1]吉林大学建设工程学院,长春130026 [2]东北电力大学电气工程学院,吉林省吉林市132012

出　　处：《吉林大学学报（地球科学版）》2011年第3期799-804,共6页Journal of Jilin University：Earth Science Edition

基　　金：高等学校博士学科点专项科研基金项目(98018706)

摘　　要：根据采空区路面塌陷数据的特性,提出了基于经验模态分解(EMD)和加权最小二乘支持向量机(WLS-SVM)预测采空区地面塌陷的新方法,并将其应用于吉林省长平高速公路因刘房子煤矿开采而引起的塌陷预测中。对实测的塌陷数据首先利用三次样条插值得到平滑的信号曲线,然后用EMD对插值后的信号进行时空滤波降噪处理,得到反映塌陷趋势的剩余分量,最后将其馈入到WLS-SVM模型完成预测。预测给出了采空区塌陷的中长期预测结果,得到塌陷区的最终塌陷值为174.34 cm,预测结果与实际监测数据平均偏差约1.06%。对长平高速公路下伏采空区段的实测数据进行分析,并与最小二乘支持向量机(LS-SVM)和BP神经网络预测结果进行了对比。结果表明:基于EMD和WLS-SRM的采空区地面塌陷预测方法具有更高的预测精度和广泛的适用性。According to the characteristic of the road collapse data in mined-out area,a new method based on empirical mode decomposition（EMD） and weighted least squares support vector machines（WLS-SVM） has been put forward to forecast the ground subsidence and applied into the coal-mining-induced collapse prediction of Changping high way in Jilin Province.Three steps has been adopted to deal with the measured collapse data： Firstly,to gain the smooth signal curve by cubic spline interpolation method.Secondly,to use EMD to deal with the time-space filtering and noise reduction to obtain the residual component which shows the collapse trend.At last,to input the dealt data to WLS-SVM model and then get the result of collapse prediction.The final collapse value has been predicted to be 174.34 cm,the average deviation of the final collapse prediction results using WLS-SVM is about 1.06%.Comparing the measured data in situ and the predicted data by WLS-SVM and BP neural network,the result shows that the suggested method has relatively high forecast accuracy and can be applied widely.

关 键 词：三次样条插值 经验模态分解 加权最小二乘支持向量机 采空区 塌陷 

分 类 号：P642.26[天文地球—工程地质学]