Estimating Model Parameters of Conditioned Soils by using artificial network

摘要

The parameter identification of nonlinearconstitutive model of soil mass is based on an inverseanalysis procedure, which consists of minimizing theobjective function representing the difference between theexperimental data and the calculated data of themechanical model. The ill-poseness of inverse problem isdiscussed. The classical gradient-based optimizationalgorithm for parameter identification is also investigated.Neural network models are developed for estimating modelparameters of conditioned soils in EBP shield. The weightsof neural network are trained by using theLevenberg-Marquardt approximation which has a fastconvergent ability. The parameter identification resultsillustrate that the proposed neural network has not onlyhigher computing efficiency but also better identificationaccuracy. The results from the model are compared withsimulated observations. The models are found to have goodpredictive ability and are expected to be very useful forestimating model parameters of conditioned soils in EBPshield.