Research Of The TBM Excavation Efficiency Prediction And Rock Classification Based On The PSO-SVR Algorithm

Tunnel construction by TBM is fast and efficient, especially in the extra-long tunnel project, the economic benefits of TBM construction are more significant. In practical engineering, in order to carry out feasibility studies, economic utility assessment and risk control, the prediction of TBM excavation efficiency indicators are needed. On the basis of the prediction, the surrounding rock class can be determined. Among the factors affecting the efficiency of the excavation, the surrounding rock geological conditions are the main external factors, surrounding rock classification in TBM construction should focus on rock boreability.In order to establish a reliable TBM performance prediction model and predict the efficiency, the support vector regression(SVR) was optimized by particle swarm optimization(PSO) algorithm, by the means of analyzing the factors that affect the efficiency, three parameters of surrounding rock—uniaxial compressive strength(UCS), the distance between the planes of weakness(DPW), the angle between plane of weakness and TBM-driven direction.(α), and driving efficiency indicators—field penetration index(FPI) were identified, as input and output parameters of the model. On the basis of the algorithm and parameters, the TBM performance efficiency prediction model was established. Meanwhile, the PSO-SVR model was compared with other models based on different theories. On the basis of the prediction model, according to the correlation between the parameters of surrounding rock and the field penetration index(FPI), a new rock classification method was proposed for the TBM construction conditions. The main achievements are as follows:(1) By means of particle swarm optimization(PSO) algorithm optimizing the key parameters selection in the regression process of support vector regression(SVR) algorithm, the support vector regression algorithm can achieve optimal fitness, achieve the best regression results. Through this research, further validate the progressiveness of the PSO-SVR joint algorithm. The method also can be applied to deal with other problems in the regression prediction in the future.(2) Compared with the linear regression, nonlinear regression, neural network theory the prediction accuracy of PSO-SVR model was the highest. The accuracy of the PSO-SVR model had improved greatly than other theories.(3) According to the correlation between the field penetration index(FPI) and the rock boreability, the surrounding rock class figure was plotted on the basis of the PSO-SVR model. The FPI value could be estimated with the rock parameters and machine parameters, and then the rock class could be determined in the rock class figure, further, the rock boreability could be estimated.