A Data-driven Approach For Predicting The Load Of A Tunnel Boring Machine And Parameter Analysis

The Tunnel Boring Machine is a systematic,intelligent and factory-built high-efficiency tunnel excavation machine.It applies to construct the long-distance tunnels and can complete the processes,such as excavation,slag transportation,ventilation,and dust removal,guiding control,supporting lining,water electricity and materials supply.A load of tunnel boring machines(TBMs)is crucial for the life span of disc cutters,cost forecast,the optimization of the TBM design as well as the guidance parameter selection.To realize the load prediction,the load model is established by the current method,which are the empirical theory formula,the rock breaking theory of the tool and the equivalent modeling method.However,the value information contained in the data recorded on the site during the TBM construction process is ignored.The method of establishing the load model based on data analysis is lacking.Therefore,it is necessary to mine the value information contained in the construction data and establish a high-precision load prediction model based on the data.Taking a single open a single pair of horizontal support TBM as the research object,the data-driven method is used to establish the tunneling load model.The model prediction performance of different modeling intervals and the parameter sensitivity of participating modeling are studied.Comparing the accuracy of the load prediction model with different parameters,a high-precision load prediction model is obtained.Compared with the measured results,the validity of the model prediction is verified.Basic theory and modeling ideas are provided to reduce the dimension of model input parameters and improve the prediction accuracy.The main research of this paper is as follows:(1)The surrogate model technology and important parameter identification technology are combined to establish the load model corresponding to different tunneling segments.Comparing with the prediction results of different models on the same prediction sections,the optimal model is determined.The method of establishing different load prediction models is analyzed.By increasing the sample data of the modeling sections,a predictive load model with continuous predictive load capacity and high precision is established to predict the load.(2)Establishing a load model of the Multi-Dimensional parameters and analyzing the parameter importance are discussed.Based on the analysis of the above modeling process,the input dimension of the model and the data sample are increased.Using machine learning to build load prediction models adapted to multiple geological types.Further,analyzing importance of the parameters and setting the parameter importance threshold to analyze the accuracy of the load model corresponding to different thresholds.So,involved in the modeling parameters are identified,which provides a reference for the adjustment of models and improving the modeling efficiency.(3)The key parameters of the load model are identified based on the parameter coefficients of the kriging model.The corresponding parameter coefficient values are obtained by analyzing the parameters of the data samples.And it is converted into the parameter weight ratio to realize the identification of the influence parameters of the load.Using the Sobol sensitivity analysis method to obtain the total sensitivity of the input parameters of the test function.At the same time,the weighting proportion of the corresponding parameters is obtained by using the kriging model parameter coefficient analysis method for the test function.The results of the two methods are compared to verify the correctness of the method.