A new hybrid grey wolf optimizer-feature weighted-multiple kernel-support vector regression technique to predict TBM performance

Abstract

Full-face tunnel boring machine (TBM) is a modern and efficient tunnel construction equipment. A reliable and accurate TBM performance (like penetration rate, PR) prediction can reduce the cost and help to select the appropriate construction method. Therefore, this study introduces a new hybrid intelligence technique, i.e., grey wolf optimizer-feature weighted-multiple kernel-support vector regression (GWO-FW-MKL-SVR) to predict TBM PR. For this purpose, a tunnel in China was selected as a case study and the most important parameters on TBM performance, i.e., chamber earth pressure, total thrust, cutterhead torque, cutterhead speed, cohesion, internal friction angle, compression modulus, the ratio of boulder, uniaxial compressive strength and rock quality designation, were measured and considered as model inputs. To show the capability of the GWO-FW-MKL-SVR model, three models including biogeography-based optimization (BBO)-FW-MKL-SVR, MKL-SVR, and SVR were also proposed to predict the TBM PR. To select the best predictive models, some performance indices, i.e., coefficient of determination (R2), root mean square error (RMSE) and variance accounted for (VAF) were considered and calculated. The obtained results showed that the GWO-FW-MKL-SVR model receives the highest accuracy in predicting the TBM PR for both train and test stages. R2 values of 0.946 and 0.894, for train and test stages of the GWO-FW-MKL-SVR model, respectively, confirmed that this new hybrid model is considered as a powerful, applicable and simple technique in predicting the TBM PR. By performing feature weight analysis, it was found that the effects of the uniaxial compressive strength, rock quality designation and cutterhead speed features were higher than the other input parameters on the TBM PR.