Dear Mohamed mustafa Mohamed iqbal,

I think the mixed method certainly presents more advantages than the singular techniques. However the choice of a mixed method depends on the representation of the system as well as whether it is continuous or discrete ...

Here are some attached files and links I hope their readings will help you make the right choice.

A Novel Method of Order Reduction for Interval System | Open Access ...

https://www.rroij.com/.../a-novel-method-of-order-reduction-forin...

Model order reduction using Fuzzy C-Means clustering - Apr 15, 2014

journals.sagepub.com/doi/abs/10.../0142331214528968?legid... 

Mixed Methods: Combining Expert Interviews, Cross-Impact Analysis ...

www.ejbrm.com/issue/download.html?idArticle=273

Best regards

 A reduced model is usually restricted to the context (operation parameters) under which it was formed. The final test normally involves verifying low white residuals on data NOT used to find the model, and in the same context. Since a power system is highly non-linear, I would choose criteria based on normalized mean square error (NMSE) since rms of residuals only apply in linear systems. Check both the criteria that trade off number of parameters and NMSE to choose the best case, together with necessary conditions of residuals with zero mean and 'whiteness'.

Dear Mohamed-Mourad Lafifi,

Thanks for your reply. Obviously the reduction technique will be based on the system. Is there any specific technique for continuous / discrete system. All the attached materials / link are useful.

Thanks.

Dear Zeashan Khan, 

Thanks for your description on how to use MATLAB for reduction.

Dear Henrietta L Galiana,

Thanks for your suggestions wr.t. linear and non-linear system.