In the fuzzy system,once we fix the input variables and corresponding to MFs, are the rules of the fuzzy system been fixed?For example ,threre are 2 input variables and corresponding to 4 MFs in each input. So how many rules in total, 8 or 16? And I have some findings in a paper of ANFIS presented in the picture? I want to know why is that? Thanks a lot!!!
hello, LiTao, one input has 4 member functions, thus the class number is 4. If you have 2 inputs, then the logical rule should be 4*4 = 16
It depends, more rules computer operation is slow. less rules, computer operation is quickly.
Well, first thing is that you can have a maximum no of rules corresponding to the permutation of the no of variables (i.e input and output), however, much more important is the "expert knowledge" that gives rise to a "rule base". you need to refer to literature, theory and decide what are the possibilities and what must happen when such a combination of inputs occur, that is, now it is up to you how to tune your FIS.
For ex:- if I'm designing a PID controller, whenever P increases, I will try to minimise the output so that the system shifts towards stability as higher P makes a system unstable. In the same way, you have to learn what "cause" has what "effect" and supply the rules accordingly. Let me know if it solves your issue. Thanks.
Initially you should consider 16 rules. But the main issue is: How many MF's do we need to represent each input variable ?? If you have data, a clustering procedure should be applied in this case in order to determine a suitable number of cluster related to each input. Each cluster can be represented by a MF.
@Cristiano Fontes thank you! I know that. I have discussed the problem with my tutor . And the no. of rules is not strictly determined by no. of mfs and input variables, namely, there may be more rules (i.e 2 mfs and 2 iinput variables generate at least 4 rules or more) Now I want to know using ANFIS function to train a liner model in MATLAB ,how can I confirm the final output function .According to the MATLAB output ,I only can obtain some structure info. including input, output ,rules ,etc. Hope you can give me some advice,thanks!
@Avijit Ashe ·Thanks a lot ! But I still cannot solve the problem that I do not know how to train a liner model usinig a function of ANFIS in MATLAB.So can you give me some detail suggestions? thanks again!
@Ray-Hwa Wong and @Wenyu Zhang thank you all ! But i am still confused about it.May I have more details?
it could be deduced that the higher the number
of rules the more effective the controller as justified by
the performance indicators.
You can see the publications of https://www.researchgate.net/profile/Serhiy_Shtovba and https://www.researchgate.net/profile/Alexander_Rotshtein or ask
them personally . They are very strong specialists in fuzzy logic.
Plot your data using bar chart, the # of the steps in the chart represents # MSF. Say 3 for the first input and 4 for the second input. The # of rules = 3 multiply by 4.
Ex : For a input if we have N membership
The maximum amount of Rules can be developed is = 2^N
- Badges
- Science method
- Similar topics
- Methodology
- Crystallography
- X-ray Diffraction