基于GWO-ELM模型的深基坑开挖变形预测研究

Prediction of deep excavation deformation based on GWO-ELM model

为有效控制和预测深基坑开挖引起的周围地表沉降变形,以保定市汽车科技产业园深基坑工程为依托,使用MIDAS GTS NX软件对实际基坑工程施工过程进行模拟,并将实际值与模拟值进行对比,验证模型的准确性。并且使用灰狼优化算法(GWO)优化了极限学习机(ELM)神经网络中的输入权重和隐藏层阈值,建立了GWO-ELM深基坑开挖变形预测模型。以有限元模型中土钉数量、开挖深度、周围建筑物沉降等因素作为预测模型的输入因子,以有限元模型中监测点DB-2地表沉降作为预测模型的输出因子。将GWO-ELM模型预测值与ELM模型预测值对比分析。结果表明:通过有限元软件提取地表沉降等数据,可以对深基坑地表沉降实现超前预测;使用灰狼优化算法对极限学习机神经网络中输入权重和阈值优化,可以提高预测模型精度;经过实际工程验证,GWO-ELM模型平均绝对误差为0.26145,均方误差为0.31258,R 2为0.98725,均优于ELM模型。

In order to effectively control and predict the surrounding surface subsidence deformation caused by deep foundation pit excavation,the deep foundation pit project of Baoding Automobile Science and Technology Industrial Park is taken as the basis.MIDAS GTS NX software was used to simulate the actual foundation pit construction process,and the actual value was compared with the simulated value to verify the accuracy of the model.In addition,Grey Wolf optimization algorithm(GWO)was used to optimize the input weight and hidden layer threshold of extreme learning machine(ELM)neural network,and the prediction model of GWO-ELM deep foundation pit excavation deformation was established.The number of soil nails,excavation depth and settlement of surrounding buildings in the finite element model were taken as input factors of the prediction model,and the surface settlement of monitoring point DB-2 in the finite element model was taken as output factors of the prediction model.The predicted value of GGO-ELM model was compared with that of ELM model.The results show that:The surface settlement and other data extracted by finite element software can achieve super prediction of deep foundation pit surface settlement.Using Grey Wolf optimization algorithm to optimize the input weight and threshold in the extreme learning machine neural network can improve the accuracy of the prediction model.Through actual engineering verification,the average absolute error of the GGO-ELM model is 0.26145,the mean square error is 0.31258,and the R 2 is 0.98725,all of which are superior to the ELM model.