Actually, the ANFIS model is created using your training data while the testing data is used to validate the created model. The lower the RMSE, the better the performance of the model.

Dear Girma,

First it depends on the representativeness of the data with process, in other hand the training data phase is also important to obtain a satisfy results.

For more details and information about this subject, i suggest you to see links on topic.

-An adaptive neuro fuzzy model for estimating the reliability of ...

https://www.sciencedirect.com/science/article/.../S22108327140001...

-Comparison of adaptive neuro-fuzzy inference system (ANFIS) and ...

https://www.sciencedirect.com/science/article/.../S13504533163015...

-Comparison of adaptive neuro-fuzzy inference system and artificial ...

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3776083/

-Proton exchange membrane fuel cell degradation prediction ... - Hal

https://hal.archives-ouvertes.fr/hal-01050920/document

-Adaptive Neuro-Fuzzy Intelligent Controller via Emotional Learning for ...

https://pdfs.semanticscholar.org/.../92c55b0ea6470210f2fa0d6f65...

-Neuro-Adaptive Learning and ANFIS - MATLAB & Simulink - MathWorks

https://www.mathworks.com › ... › Adaptive Neuro-Fuzzy Modeling

-A Modified Neuro-Fuzzy System Using Metaheuristic Approaches for ...

https://www.intechopen.com/.../a-modified-neuro-fuzzy-system-usi...

Best regards

Hi Girma Kassa, let's get to the real intention behind your question.

From your Project, you stated that to use the PI controller data for ANFIS training so that it can replace the fractional-order PI controller. If you already have a working fractional-order PI controller in the first place, why do you need a fuzzy controller then? I wonder... 🤔

Are you investigating if a fuzzy controller performs "better" than the fractional-order PI controller? 💪

Dear Yew-Chung Chak ,

Fractional order PI controller has fixed gain and integration order,the performance of this controller diminishes under certain uncertainties like parameter variation,...But,once ANFIS is trained using i/o data from Fractional order PI controller,it can generalize from unseen data even though uncertainties occur .

Hi Girma Kassa, thanks for your reply. It is certainly interesting to find out what the ANFIS is capable of.

If the ANFIS controller is well-trained (without enhancement or modification), isn't its performance similar to the fractional-order PI controller?

If the I/O data of fractional-order PI controller is deterministic, and the ANFIS is well-trained, will it become more robust in the presence of uncertainties?