Hi, according to my small experience in using PSO methods I have found that, for a given algorithm, the efficiency to converge to the global optimum strongly depends on the tuning of the method (number of particles on the initial population, etc..) and also on the problem to be solved. Thus, I think it may be very difficult to say that a given method/algorithm is better than another one. You can certainly find on the web test cases and papers containing comparisons between different PSO variants.

Dear Lemuel Carlos Ramos Arzola,

I suggest you to see links and attached files in topics .

- Handbook of Research on Modern Optimization Algorithms and ...

https://books.google.dz/books?isbn=1466696451

- Multidimensional Particle Swarm Optimization for Machine Learning ...

https://books.google.dz/books?isbn=3642378463

- Issues in Computer Science and Theory: 2012 Edition

https://books.google.dz/books?isbn=1481647032

- Advances in Swarm Intelligence: 8th International Conference, ICSI ...

https://books.google.dz/books?isbn=3319618245

- A Review of Particle Swarm Optimization Methods used for Multimodal ...

delta.cs.cinvestav.mx/~ccoello/chapters/chapter-barrera-final.pdf.gz

Best regards

Dear friend

You should follow my research paper. U have same topic with me. We will discuss ur problems in the end of November. My papers will solve ur problem which mention u above.

With best wishes

Sharandeep singh

Thanks for your answer.

Dear Lemuel Carlos Ramos Arzola,

I did work for a little while with PSO but I didn't carry out an in-depth investigation about the performances of the different variants of the algorithm. In fact, the basic implementation of PSO was enough in my case.

Anyway, one of the main classifications of the existing PSO variants is whether it is a global or a local search. Generally speaking, a global search is accomplished by dividing the swarm into several smaller ones that share information every now and then (something like the island structure of some other meta-heuristic algorithm that you can find in the literature, as ESA's PyGMO). Of course, the frequency the swarms interact and the amount of information they share are two heuristics that can affect the quality of your final solution, in terms of speed of convergence and global characteristic of the solution, for instance.

In my opinion, a good test bed for assessing the performances of the algorithm you want is the Congress of Evolutionary Computation (CEC). They carry out an annual competition in which they test meta-heuristic algorithms on multidimensional multimodal mathematical problems. In particular, during the 2005 edition of it (CEC 2005: http://web.mysites.ntu.edu.sg/epnsugan/PublicSite/Shared%20Documents/Forms/AllItems.aspx?RootFolder=%2fepnsugan%2fPublicSite%2fShared%20Documents%2fCEC2005&FolderCTID=&View=%7bDAF31868-97D8-4779-AE49-9CEC4DC3F310%7d ), two variants of PSO participated in the competition. You can read the related papers, which are the following:

(1) Fatih Tasgetiren, M., Liang, Y. C., Gencyilmaz, G. and Eker, I., “Global Optimization of Continuous Functions Using Particle Swarm Optimization”, IEEE Congress on Evolutionary Computation (Special Session on Real-Parameter Optimization), IEEE Paper 519, Edinburgh, Scotland, UK, 2005.

(2) Liang, J. J. and Suganthan, P. N., “Dynamic Multi-Swarm Particle Swarm Optimizer with Local Search”, IEEE Congress on Evolutionary Computation (Special Session on Real-Parameter Optimization), IEEE Paper 620, Edinburgh, Scotland, UK, 2005.

Hope this is useful for you.

Cheers,

Alessandro

Hi Lemuel,

Short answer:

iSRPSO/DMSPSO

Long answer:

we did a comprehensive benchmark on PSO variants, and the results are summarised here Conference Paper On the Performance of Particle Swarm Optimization Algorithms...

I recommend my NSC-PSO method (see my article "NSC-PSO, a Novel PSO Variant without Speeds and Coefficients") where I combine ideas from PSO and ABC and there are no parameters to tune. In the article I compare NSC-PSO with SPSO 2007, which was the standard at that time, and NSC-PSO proved much better.

If it is possible you should apply divide-and-conquer techniques first then apply the PSO while testing the parameters for the best combination. Branch and Bound is quite useful in some cases.

Try fruit fly optimization algorithm Or Hybridize between any two swarm optimization algorithms

I would recommend multi-swarm systems:

Conference Paper Multi-swarm hybrid for multi-modal optimization

Conference Paper WoSP: A Multi-Optima Particle Swarm Algorithm

Conference Paper An Analysis of Locust Swarms on Large Scale Global Optimizat...

Conference Paper An Analysis of Sub-swarms in Multi-swarm Systems

Visit link for Matlab Code

https://youtu.be/OQ3T575sbI8

http://dg-algorithm.blogspot.com/

All the answers are great

... and very likely none will lead to an optimum.

There are global optimisation solvers, too, and for free:

https://se.mathworks.com/products/global-optimization.html

It does not hurt to try, does it? Try at least some of the low-dimensional test cases that can be found on the web, and let these global optimisation tools compete with the latest metaheuristics - that might be an eye-opener for us all! :-)

I do not have the tools, but if the ones who read this can take a few test cases each, and let them compete fairly, it might be a nice contribution!

I am not sure that I want to be a moderator in this, but if you do allow the two "principles" - i.e., global optimisation versus metaheuristics - I can perhaps make notes and look at the outcomes, unless someone else volunteers.

Of course, this pretty much means that we have to use instances where we know the optimum and the optimal value of the objective function, but I do think we can find at least that info on the site above, and if there are other trusted sources of problems and known optima, by all means run your methods.

This might be fun, and worth-while.

Michael

Lemuel Carlos Ramos Arzola

Hi bro

I attached some variants of PSO here for you