Improving the speed performance of an Evolutionary Algorithm by a second-order cost function approximation

In the present paper, dynamic identification problem of a FE structure with unknown parameters is solved by global search method. Response surface methodology is introduced in Differential Evolution algorithm to improve the performance of the algorithm. Differential evolution (DE) is an evolutionary algorithm where N different vectors collecting the parameters of the system are chosen randomly or by adding weighted differences between vectors obtained from two populations. In the modified algorithm, the new parameter vector is defined as the minimum of a second-order polynomial surface, approximating the cost function. Performance in term of speed rate is strongly improved by introducing the second-order approximation; nevertheless, robustness of DE algorithm for global minimum search of cost function is preserved, since multiple search points are used simultaneously. A numerical examples is presented, concerning identification of mechanical parameters of a steel truss girder bridge with unknown values of masses and stiffnesses of bracing and bearing.