Abstract: To obtain more effective factors affecting the surface settlement data for machine learning algorithm of surface settlement prediction, the support vector regression machine based on differential evolution algorithm (DESVR) of machine learning methods is studied to establish a surface settlement prediction method, combining with the influence range of surface settlement during shield tunnel construction stage, as well as multivariate time series data characteristics of surface subsidence influencing factors within the scope of the influence range. The prediction method of surface settlement is applied in Changzhou rail transit line 1 project, and the results show that the proposed method has higher prediction accuracy and more stable prediction effect compared to the traditional machine learning prediction method.

Key words: shield tunneling, surface settlement prediction, influence range, covariance matrix, support vector regression machine