Dear Aschkan Omidvar,

I would like to recommend you the following article:

M. N. Alam, B. Das, V. Pant, A comparative study of metaheuristic optimization approaches for directional overcurrent relays coordination, Electric Power Systems Research 128 (2015) 39-52.

I hope, it will be helpful to you..

Thanks

Dear Aschkan Omidvar,

the best way is to use an automated tuning approach on a pre-specified problem by get the range for each of the tuning parameters [min , max] value then run a loop and record the parameters values and results. the best optimum result is given the best parameters values.

this is the best available approach now. however, it time consuming in a large space problems. 

best of luck.

 Hello!

One interesting idea is to use a meta-metaheuristic approach for tuning parameters of metaheuristics. The section 1.6 of the book "Metaheuristics: From Design to Implementation" (see link) introduces that concept.

Good luck. Best regards!

https://books.google.com.br/books?id=SIsa6zi5XV8C&pg=PA55&lpg=PA55&dq=metametaheuristic&source=bl&ots=-9TMtOfrAt&sig=CiU5S7Iif5xJDm0yfLhVshFhn3k&hl=es&sa=X&ved=0CEcQ6AEwBWoVChMI9tmbk-mNyQIVgiCQCh1rGAyk#v=onepage&q=metametaheuristic&f=false

you can use self adaptive tuning methods in which the parameters get tuned itself during every iteration. many papers are available for self tuning of parameters for all algorithms. please refer to that.

There are two different strategies for parameter tuning: the off-line parameter initialization (or meta-optimization) and the online parameter tuning strategy. In off-line parameter initialization, the values of different parameters are fixed before the execution of the metaheuristic, whereas in the online approach, the parameters are controlled and updated dynamically or adaptively during the execution of the metaheuristic.

I recommend taguchi method due to its ease of implementation. The following reference may help you: Talbi, El-Ghazali. Metaheuristics: from design to implementation. Vol. 74. John Wiley & Sons, 2009.  

I endorse Ashkan's comment, but I draw a different conclusion :P

Try SMAC, ParamILS or irace for offline tuning. So far I've only used irace, so it's the only one I can help you with. I prepared this tutorial a while ago for a tuning workshop. Maybe it can help you: http://iridia.ulb.ac.be/irace/tutorial/

Just be careful that you have a clear separation between a tuning set and a testing set.

Mahamad,

Thanks for the suggestion. However, I scanned the paper but I couldn't find any topic related to tuning the parameters. I saw parameters, such as CF, MF, CR, F and HMCR are selected based on a procedure but no discussion was I able to find on the optimality and validity of those parameters. Could you please elaborate more which section you were trying to refer me to?

Thanks!

Yazan,

thanks for the suggestion. I have two questions on your approach. first, how this method considers the interactions of parameters. For example in SA, the interaction of initial temp, goal temp and cooling factor is important in the solution quality and optimizing these parameters one at a time might not lead to an overall solution. Second, I was wondering if you have/ encountered/ reviewed any paper which used this approach.

Reynier,

thanks for your suggestion. I actually liked the concept. Maybe I would use a simple meta heuristic solver package to run a simple MH algorithm on the parameters.

Sabarinath,

thanks for the great suggestion. It sounds like an interesting approach and I'll consider it on my research.

Ehsan,

thanks for your suggestions. So far I've implemented Factorial design, RSM and in one case, Taguchi DoEs. I was curious if there are other faster and more efficient methods we can use besides ANOVA or experiments like that. 

Ashkan,

thanks for your suggestion and sufficient explanation. I keep receiving suggestions on self-tuning and online- tuning methods. Seems these methods are gaining more attention in the literature. I'm gonna consider this approach in my research. Thanks!

Leonardo,

Thanks for your suggestions, and a special thanks for providing concrete example and tools. I checked them all (except for SMAC which wasn't quite easy to search and I need to take my time to search and find the topic!) and liked their concept. I will consider them in my research.

Hello! I have the similar problem in my research. I found this software to be potentially useful: Bonesa, https://sourceforge.net/projects/tuning/. A related paper is Multi-Problem Parameter Tuning using BONESA. But I ran into a problem using this software: the visualisation module did not work. Another problem is that because of the lack of examples and docs, I have no idea how to use the multi-problem tuning.

This may be useful for your research.

From the above discussion, irace looks a promising option. I would like to try it in my research. Thanks Leonardo!

You can also consider settings-free meta-heuristics, in which a computational intelligence methodology takes care of selecting the optimal settings at run-time. For instance, multiple versions of PSO hybridized with fuzzy logic have been proposed. One example is the following paper:

Nobile M.S., Pasi G., Cazzaniga P., Besozzi D., Colombo R., Mauri G.: Proactive particles in swarm optimization: a self-tuning algorithm based on fuzzy logic , Proceedings of the 2015 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2015), Istanbul (Turkey), pp. 1–8, 2015

I know that my answer comes almost two years after asking the question, but I feel that I can contribute to the discussion. I used the Design of Experiments approach to tune Bee Colony Optimization. I used Box–Behnken design to screen the parameters, analyzed the screening results with mixed models and visually, and then tuned the parameters using RSM. I did everything with R and python packages. If anyone wants to read the paper, its accepted version is fully available on my blog.

https://rhythm-of-the-algorithm.blogspot.rs/p/blog-page_3.html

Article Is the Bee Colony Optimisation Algorithm Suitable for Contin...