Identification of structural systems subject to earthquake excitation using an advanced genetic algorithm

Within the framework of identification methodologies, so-called non-classical methods based on soft computingtechniques have shown a promising robustness and efficacy. In particular, in the last decade an increasing attentionhas been paid on biologically-inspired routines (i.e., neural networks and genetic algorithms) to identify modelscharacterized by linear as well as nonlinear behaviour. In this paper, an advanced genetic algorithm is presented for parameter identification of linear multiple-degree-of-freedom structural systems subject to earthquakes. The proposed algorithm utilizes several subpopulations and chromosomes are represented by means of real encoding.Moreover, recent developments in traditional genetic operators (crossover and mutation) are taken into account, sothat the final algorithm combines an adaptive rebirth operator, a migration strategy and a search space reductiontechnique. This method is here applied to identify a shear-type mechanical system with unknown masses andunknown stiffness from an incomplete set of measurements: in detail, the identification is carried out with referenceto a limited set of monitored acceleration responses. Numerical results show that the proposed strategy outperformsexisting genetic algorithms for structural identification and its numerical robustness can also effectively cope withnoisy signal