The choice of options depends on the identification problem. Kindly consider this excerpt from the old documentation: https://fomcon.net/pub/temp_docs/fotfid_tool.pdf
In short: What type of model would you like to obtain? You can try obtaining a model with a commensurate order, in which case the model generation panel and “fix exponents” option may be useful. You can also identify integer-order models by fixing exponents with $\alpha_k, \beta_k\in\mathbb{Z}^{+}$.
The identification process involves constrained optimization, where the objective function is based on the model’s output error (see MATLAB’s Optimization toolbox documentation for lsqnonlin). That is, the output error is minimized in the least squares sense.
