Genetic-Algorithm-Based Strategies for Dynamic Identification of Nonlinear Systems with Noise-Corrupted Response

The main objective of this paper is to investigate efficiency and correctness of different real-coded genetic algorithms and identification criteria in nonlinear system identification within the framework of non-classical identification techniques. Two conventional genetic algorithms have been used, standard genetic algorithm and microgenetic algorithm. Moreover, an advanced multispecies genetic algorithm has been proposed: it combines an adaptive rebirth operator, a migration strategy, and a search space reduction technique. Initially, a critical analysis has been conducted on these soft computing strategies to provide some guidelines for similar engineering and physical applications. Therefore, the hysteretic Bouc-Wen model has been numerically investigated to achieve three main results. First, the computational effectiveness and accuracy of the proposed strategy are checked to show that the proposed optimizer outperforms the aforementioned conventional genetic algorithms. Secondarily, a comparative study is performed to show that an improved performance can be obtained by using the Hilbert transform-based acceleration envelope as objective function in the optimization problem (instead of the pure acceleration response). Finally, system identification is conducted by making use of the proposed optimizer to verify its substantial noise-insensitive property also in the presence of high noise-to-signal ratio.