Research On Tbm Performance Prediction And Rock Mass Parameters Characterization Method Based On Machine Learning

In TBM construction of tunnel,the inability to obtain rock mass parameters in front of tunnel face in a timely and accurate manner usually leads settings of tunneling parameters to be hugely influenced by human factors and tunneling parameters not to match with the rock mass parameters,resulting in low efficiency of rock breaking and low TBM utilization,even abnormal wear of cutters and cutterhead and other disasters,which seriously affect TBM's safe and high efficient excavation.Over the years,many scholars have devoted themselves to the TBM performance prediction,but the prediction method is usually focused on simple multiple linear regression and accuracy and generalization performance of most of the models are difficult to meet the engineering requirements,resulting in lack of accurate relationship models between rock mass parameters and tunneling parameters in TBM construction process.To address the problems and satisfy requirements mentioned above,in view of the advantages of machine learning method in revealing the implicit laws of multi-feature data sets,this paper proposes the idea to establish data models based on machine learning algorithm,and use the models to predict rock mass parameters and TBM performance parameters during mining.In this respect,this study established database of rock mass parameters and TBM performance parameters relying on water supply project for Songhua River.Rock mass parameters characterization and TBM performance parameters prediction are realized with the support of vector machine regression algorithm.The main contents studied in this paper are as follows:(1)Optimum selection of machine learning methods for TBM tunneling performance prediction and characterization of rock mass parameters.Based on the characteristics of built sample dataset of rock mass parameters and tunneling performance parameters,the research tentatively selected six machine learning algorithms including BP neural network,support vector machine and tree regression to compare and analyze their features.Through analysis,it's concluded that compared with other algorithms,the support vector machine algorithm has better performance for medium and small scale data samples and has better ability to obtain the global optimal solution;at the same time,support vector machine regression algorithm can effectively avoid over-fitting and has strong generalization ability,and therefore it is more suitable for the study.(2)Establishment and optimization of rock mass parameters characterization models using TBM main tunneling parameters and predication of rock mass parameters with rock mass parameters models.Respectively based on the BP neural network,tree regression and support vector machine regression algorithm,data mining models of seven rock parameters including the joint spacing(DPW),uniaxial compressive strength(UCS),brazilian tensile strength(BTS),joint direction,brittleness index(Bi),rock hardness(HS)and rock quality index(RQD)using six TBM parameters including total thrust,cutterhead torque,penetration,etc.were built and excavation predication was done.The results show that the prediction effect of the support vector machine regression model is the best,and the prediction results of the above six rock parameters(brittle index)are consistent with the actual value trend with the average eMAPE value of 18.58%.The parameters of rock mass in limestone geological section and limestone,tuffaceous sandstone mixed geological section of Yinsong water supply project 4th bid section were collected and support vector machine regression models were built to do predication.The average eMAPEe values obtained are 19.01%and 20.83%respectively,which shows that the models have good prediction performance.(3)Establishment and optimization of prediction model of the main TBM performance parameters using the main rock mass parameters and predication of tunneling parameters.Respectively based on BP neural network,tree regression and support vector machine regression algorithm,data mining models of the five TBM operating parameters including driving speed,thrust,torque,penetration and penetration index using the seven rock mass parameters described in(2)were built and excavation predication was done.The results show that the prediction effect of the support vector machine regression model is the best.The prediction results of the five TBM performance parameters above are consistent with the trend of actual values and the average eMAPE is 21.94%.The parameters of rock mass in limestone geological section and limestone,tuffaceous sandstone mixed geological section of Yinsong water supply project 4th bid section were collected and support vector machine regression models were built to do predication.The average eMAPE values are 19.50%and 19.36%respectively,which shows that the models have good prediction performance.(4)Engineering application in Yinsong water supply project 4th bid section and verification of generalization of the models.Rock mass parameters and TBM performance parameters of 30 test areas of the project were predicated using the aforementioned support vector machine regression model.The prediction results are close to the results verified by the model,and average eMAPE values are 20.42%and 20.56%respectively.It is proved that the support vector machine regression model has good generalization ability.