Computer-Aided Civil and Infrastructure Engineering 26 (2011) 92–110 Modified Genetic Algorithm for the Dynamic Identification of Structural Systems Using Incomplete Measurements Giuseppe Carlo Marano ∗ Department of Environmental Engineering and Sustainable Development, Technical University of Bari, viale del Turismo 10, 74100 Taranto, Italy & Giuseppe Quaranta & Giorgio Monti Department of Structural Engineering and Geotechnics, Sapienza University of Rome, via A. Gramsci 53, 00197 Roma, Italy Abstract: A modified real-coded genetic algorithm to identify the parameters of large structural systems sub- ject to the dynamic loads is presented in this article. The proposed algorithm utilizes several subpopulations and a migration operator with a ring topology is pe- riodically performed to allow the interaction between them. For each subpopulation, a specialized medley of recent genetic operators (crossover and mutation) has been adopted and is briefly discussed. The final algorithm includes a novel operator based on the auto-adaptive asexual reproduction of the best individual in the cur- rent subpopulation. This latter is introduced to avoid a long stagnation at the start of the evolutionary process due to insufficient exploration as well as to attempt an improved local exploration around the current best solu- tion at the end of the search. Moreover, a search space re- duction technique is performed to improve, both conver- gence speed and final accuracy, allowing a genetic-based search within a reduced region of the initial feasible do- main. This numerical technique has been used to iden- tify two shear-type mechanical systems with 10 and 30 degrees-of-freedom, assuming as unknown parameters the mass, the stiffness, and the damping coefficients. The ∗ To whom correspondence should be addressed. E-mail: gmarano@ poliba.it. identification will be conducted starting from some noisy acceleration signals to verify, both the computational ef- fectiveness and the accuracy of the proposed optimizer in presence of high noise-to-signal ratio. A critical and detailed analysis of the results is presented to investigate the inner work of the optimizer. Finally, its performances are examined and compared to the most recent results documented in the current literature to demonstrate the numerical competitiveness of the proposed strategy. 1 INTRODUCTION Although a wide number of mechanical system identifi- cation theories have been suggested in the past years, this is still an active research field, being a vital de- mand in many different engineering areas, such as struc- tural health monitoring and structural control. A central question regarding the identification of large structural systems is the efficiency and the robustness of the adopted strategy when the system is equipped with few measurement points (e.g., sensors). This is a very impor- tant task, because economical and practical constraints impose a reduction of the total amount of sensors to install on the structures, as compared to the available C  2010 Computer-Aided Civil and Infrastructure Engineering. DOI: 10.1111/j.1467-8667.2010.00659.x