I'm trying to use the datafit function, following step by step the procedure indicated in corresponding Scilab help. If I create a data pair collection like in the example (adding noise to a known function) the parameterrs obtained (p and err in the example) can fit "perfectly" a curve proppoused like a model, but, if I have a set of pairs obtained experimentaly (with a behaivour corresponding to a model) I can't obtain values to p an the error is very high. What could be te problem wiith two ways to generate the data?
Best regards, and thanks for your advices
If I understand your question correctly, you want to know why the fit is better for simulated data against the experimentally obtained real data. If so is the case, in my opinion, it has to do with the order of the model. You see, the simulated data is obtained from a model whose order is exactly known. However, in case of fitting a curve on data obtained through experiment, the model is an "approximation" of the real system. The real system may be simpler or many times complicated or require additional parameters like delay etc.
Secondly, the experimentally obtained data could be too noisy with some datapoints deviating too much from the steady-state value.