A hybrid multi-objective algorithm to predict the characteristics of soil profiles from seismic ground motion records

Abstract

The underlying goal of this study is to present an efficient algorithm to identify soil parameters such as thicknesses, shear wave velocities, damping and others parameters of subsurface layers, and site amplification characteristics (natural frequencies, peak amplitudes) from a given pair of seismic records. It is a hybrid procedure combining the stochastic genetic algorithms (GAs) optimization method, to find a point close to the global optimum in the global search phase, and a gradient based local determinist method (Levenberg-Marquardt: LM), to refine the solution. To improve the performance of the global search phase, a multi-objective optimization algorithm is used to minimize the errors between some characteristics of the theoretical amplification function and the experimental one of vertical array records. The weighted sum method which combines the weighted objectives into a single objective function is used to solve the optimization problem. The efficiency of the present algorithm is proven by several examples. Results show that the scheme works well and the curve fitting was always satisfying. Also, the proposed procedure leads to good approximations, requiring a lower computational effort, yet with good rates of convergence. Moreover, neither the growing number of parameters nor the vastness of the search space reduces the efficiency of the algorithm in predicting the characteristics of soil profiles and site amplification commonly required in seismic risk mitigation.