Study On Prediction Of Settlements Of TBM Tunnel Construction In Guangzhou By Using SVR Algorithm

With the rapid development of underground in modern city, TBM(Tunnel Boring Machine) has been used widely in tunnel construction because of its high formation adaptability and good safety performance etc. In practical tunnel project, however, the ground settlement, which caused by inappropriate construction, will lead to some corresponding safety problems. To deal with these, taking Guangzhou Metro Line 13 as a background, a study on surface settlement prediction by using Support Vector Regression(SVR) learning algorithm is introduced in this paper. The main content as follows:1. The engineering parameters have been divided into TBM parameters and hydrogeological parameters. The influences of different parameters on surface settlement have been discussed by comparing the practical data, and the main factors are determined eventually.2. By comparing empirical formula proposed by Peck and theoretical formula with practical data respectively, it is proved that empirical formula performs better than other formulas. In addition, the transverse distribution law of settlement during this construction interval is also proposed.3. Unlike the traditional artificial neural network(ANN) which will over fit easily, SVR algorithm are among the best(and many believe is indeed the best) "off-the-shelf" supervised learning algorithm, so that SVR is proposed to predict the surface settlement in this paper. SVR algorithm will be introduced by theoretical derivation firstly, and Relative Error and Mean Square Error(MSE) are considered as two indexes to present and validate the precision of results. After comparing the influences of different parameters on SVR prediction model, it can be concluded that TBM parameters have a great influence on settlement prediction.4. Finally, taking Back Propagation(BP) Neural Network to compare with SVR model, and the results show that SVR model has a better estimate capacity and a higher predicting precision than the former one.