A hierarchical soft computing model for parameter estimation of curve fitting problems

The aim of this paper is to present an alternative solution model to estimate the coefficients of large-scaled linear and nonlinear real-life problems due to the fact that least squares and least median squares parameter estimators have some drawbacks when including so many input variables or increased size of the real-world problems. The study presents a hierarchical soft computing model (SOFTC) that consists of three stages. The first stage constitutes a real-valued breeder genetic algorithm (RVBGA). The second stage is constructing a simulated annealing (SA) algorithm in which the best parameter estimation of the RVBGA is selected as its initial point. The third stage is developing a hierarchical soft computing model by using fuzzy recombination method. SOFTC optimizes the best parameter estimations of this algorithms and it provides a trust region for parameter estimation. Three test problems, one of which is linear and others are nonlinear, are used to examine robustness of proposed models. SOFTC, RVBGA_SA and RVBGA algorithms performed the best parameter estimations, respectively, for the three test problems. The results which are discussed in detail are promising for future usage of these algorithms.