Research On TBM Tunneling Efficiency Prediction Model In Mixed Ground Based On Machine Learning

Tunnel Boring Machine(TBM)construction method has gradually become the main construction method for underground tunnels in China due to its advantages of high quality,high construction safety,good construction environment,low construction cost and fast construction schedule.The construction efficiency of TBM is sensitive to geological conditions,especially in mixed ground,which makes it easy to have stuck tool plates,shields and large deformation of machine position.This makes rock-machine interaction a strong nonlinear feature,which makes it extremely difficult to accurately predict TBM construction efficiency and adjust TBM construction parameters in a timely manner.Therefore,establishing a reasonable and accurate prediction model for TBM penetration rate is one of the key issues to improve the TBM intelligent auxiliary driving system and enhance the TBM tunneling efficiency.This paper focuses on the prediction of TBM tunneling efficiency under mixed ground conditions,and the main innovations include:(1)On the basis of combing the influencing factors of TBM penetration rate in the literature,combined with the actual situation of TBM tunneling construction in the mixed ground of the South-North Water Transfer Project in western Anyang,the main factors affecting TBM tunneling construction in the mixed ground were analyzed through the data collected in the field,and the influence law of each factor and the fitting relationship with the penetration rate were summarized and analyzed to establish a reliable TBM tunneling construction prediction data set.(2)Based on two linear decreasing inertia weight methods to improve the traditional PSO algorithm and combined with least squares support vector machine(LSSVM),An improved PSO-LSSVM point prediction model for TBM construction efficiency in mixed ground is established.The construction efficiency prediction is carried out using engineering construction data sets and computationally compared and analyzed with the traditional PSO-LSSVM,LSSVM and SVM models.The results show that the method proposed in this paper can effectively improve the superiority-seeking capability of the PSO algorithm and the prediction effect is better than the remaining three traditional prediction models.(3)Considering the strong uncertainty of TBM construction efficiency under mixed ground,based on interval prediction theory,four different Bootstrap sampling methods are used to sample the original data set to build a new TBM data set: Bootstrap-ANN-KELM,a TBM penetration rate interval prediction model,was developed based on artificial neural network and kernel limit learning machine algorithm.The prediction results show that the model predictions established by four different methods have achieved better results,among which the Wild Bootstrap method based on Rademacher distribution achieves better accuracy at 90%,95% and 99% confidence level,making the obtained interval can completely envelop the actual measured values of TBM digging speed within the interval,and the model prediction interval is more reliable at the confidence level of 99%.