Abstract: An improved differential evolution algorithm (IDE) is proposed by introducing adaptive factor and is applied to optimizing the kernel parameter and penalty parameter of online support vector regression (OSVR). A dynamic optimal IDE-OSVR model which reflects the complex nonlinear relationship between rock and soil mass displacements and geotechnical parameters is established. The inversion of parameters can be accomplished by inputting the soil mass displacements in the IDE-OSVR model. The initial training set is designed with uniform design method and ABAQUS calculation, then the errors of successive inversed parameters are checked, and the checked sample will be added to training set if the error is greater than the predetermined threshold. Through the continuous online learning process, the precision of parameters inversion by IDE-OSVR model can be enhanced. The inversion method based on IDE-OSVR-ABAQUS is applied to an engineering example and the result is compared with those of typical methods. The comparison result shows that the IDE-OSVR-ABAQUS inversion method is very fast with high accuracy, hence is a reasonable method for the inversion of geotechnical mechanics parameters.

Key words: IDE, OSVR, ABAQUS, parameters inversion, soil mass displacement