Soft-computing-based parameter identification as the basis for prognoses of the structural behaviour of tunnels

Abstract

A parameter identification (PI) method for determination of unknown model parameters in tunnelling is presented. The PI method is based on measurement data provided by the construction site. Model parameters for finite element (FE) analyses are identified such that the results of these calculations meet the available measurements as well as possible. For the determination of the unknown parameter set, use of an artificial neural network (ANN) is proposed. The network is trained to approximate results of already performed FE simulations. A genetic algorithm (GA) uses the trained ANN to provide a prognosis for an optimal parameter set which, finally, must be assessed by an additional FE analysis. In contrast to other gradient-free methods requiring a large number of FE simulations, the proposed PI method renders back analysis of model parameters feasible even for large-scale models. Finally, the performance of this PI method as the basis for prognoses of the structural behaviour of a tunnel is demonstrated.