The Improved Fast Recursive Method and Its Application in Group Method of Data Handling

Based on the analysis of multicollinearity in FRA, an improved FRA with a novel algorithm for detecting and deleting the linearly correlated regressors is proposed. Thus excessive computation cost may be reduced and the reduction ratio is also analyzed rigorously. A novel GMDH which uses the improved FRA to estimate the coefficients in the partial description is also proposed. The experiments and analyses prove that the improved GMDH produces the constructed model exhibiting a higher level of accuracy and superior generalization capabilities in a parsimonious style