Dear Saeed Rahmati,

there are several ways. Now I list two possibilities.

1) If can produce a significant number of the desired I/O of your desired controller you can use this data with automatic tools that extract fuzzy knowledge given a prefixed number of rules. 

2) Another way that you can follow is to approximate your fuzzy controller. You can extract the above mentioned data from your large fuzzy controller and use the tools I described before.

For having an idea of these tools see my works in which I describe FuGeNeSys and the more recent GEFREX.

Good look

Marco

I use an approach which differentiate fuzzy model attributes into real and imaginary areas. This step reduces number of dimensions.

Here is my personal list of references on this topic.

Please find few links on rule reduction technique in fuzzy system..

http://link.springer.com/chapter/10.1007%2F3-540-33517-X_3

http://www.sciencedirect.com/science/article/pii/S0888613X07001028

Here are links to my RG publications on the theme:

Basic approach: https://www.researchgate.net/publication/258817515_The_control_automation_system_method_based_on_combined_implementation_of_fuzzy_logic_and_discrete_automation_systems

Example: https://www.researchgate.net/publication/264974259_The_search_engine_based_on_combined_implementation_of_fuzzy_sets_and_complex_computing

Conference Paper The search engine based on combined implementation of fuzzy ...

Dear Saeed Rahmati,

the methods I wrote before are have an interesting property: the number of rules is not tied to the power of the number of the inputs. For example, I have developed several fuzzy models with a few tens of rules (or less) with low errors and tens of inputs. Many other tool in literature do the same thing.

If your problem is the final computational time of your fuzzy model this is a great possibility.

I hope this help you

Marco

Utilizing either a cascaded fuzzy system, or a fuzzy tree (if applicable), you can significantly reduce number of rules required to run your system.  I have a few publications on the topic you could check out.

You can reduce it by using some optimization techniques such as PSO, GA etc.

using fuzzy c-means clustering