Hybrid Method for Adjusting Models of Nonlinear Regression

Resumo

Regression analysis is widely applied in many fields of science and engineering to predict any variable
that is difficult to determine. Generally, Nonlinear regression models are more complex than linear regression.
More traditional models require initial parameters to be adjusted, and the procedure for estimating these values is
not simple. In this paper, a hybrid method is proposed to meet nonlinear needs. The proposed method consists
of two steps. In the first one, two genetic algorithms (GA) were applied for automated parameter prediction. The
second step consisted of applying the Levenberg-Marquardt algorithm (GALM) to obtain the appropriate values.
The T LBO and Differential Evolution algorithms were tested to estimate the initial parameters. The model selected
for the study concerns the population analysis of individuals; and the database describes the forest inventory of the
Tectona Grandis planting in southern Minas Gerais. The results showed that both genetic algorithms were efficient
to estimate the initial parameters, with equivalent square mean error. Therefore, it is concluded that the proposed
hybrid method is effective for estimating the initial parameters.