Structure detection for nonlinear rational models using genetic algorithms

A new nonlinear rational model identification algorithm is introduced based on genetic algorithms. Compared with other rational model identification approaches, the new algorithm has two main advantages. First, this algorithm does not require a linear-in-the-parameters regression equation and, as a consequence, the severe noise problems induced by multiplying out the rational model are avoided. Second, the new algorithm provides near-optimal global parameter estimation. Unfortunately, this is balanced by an enormous computational load even when identifying models which consist of modest parameter sets. Simulated examples are included to illustrate that the new algorithm works well on systems with modest candidate term sets but can fail when applied to systems with large candidate term sets.