I think your project has the wrong schedule, if we can call it that: You are NOT required to utilize ANY metaheuristic algorithm, especially if it turns out that there are much better methodologies for the problem that you want to solve. I insist that you should *not* first start by providing a long list of algorithms, but to start by analysing the properties of your problem. Once you have done that you should design one or more methodologies to "attack" your problem. There are plenty of things to discuss as far as problem properties are concerned:

(a) Do you have an algebraic description of the problem - that is - explicit functions describing the objective function(s) and constraints?

(b) If not, how do you calculate the objective function, how long does it take, and is the calculation done with a good precision?

(c) What is the requirement on the result? Do you need a global optimal solution, or is it OK to get anything that is better than the best function value that you have at hand right now?

(d) Do you wish to solve many similar problems, with minor adjustments on problem paramaters, or just one problem instance?

You see the point? There is no purpose being narrow-minded; the best is to try several techniques; even if some fail, you will get some insight from the experience.

The efficiency of a metaheuristic (evolutionary algorithm, tabu search, PSO) is dependent on the way you have adapted it to your problem. For example, the solution coding, the reproduction operators. Have you add a local search to your algoithm. You can start by local search and after that insert your local serach into a population based algorithm.

Dear Abu,

There are few more..like

Its difficult to compare these algorithms.. They depends on the characteristics of the system. One algorithm may give better results for a particular case, but may fail to other.. There are couple of articles where they compared with different objective functions with parameters 2-10.

yeah sir, we need to share our knowledge to discover this field

I suggest that you specify the problem/model clearly first - perhaps actually posting it here. Without a clear - explicit! - formulation of the problem at hand there is no way that we can help out. You should realize that the best algorithm is not necessarily the one you have in mind: it might depend on many things, including what subroutines you have available to solve "subproblems", whether data is "crisp" (i.e., perfectly known) or known only in terms of probability distributions, and many other factors. Right now we have too little information in order to provide a guide. Throwing random algorithm names around does not help at all - as we have too little information.

And: I would never start by using a meta-heuristic; that will provide very little insight, as they are approximate - that is - they will with probability near 0 give you an optimal solution.

Dear sir Patriksson may be you are right but primarily first we need to know how many meta-heuristics are available to solve our problem then secondly we can discuss which one is best for what type of problem?You can share also your thinking here for any special type of algorithm for a specific problem so that one can benefited from your answer

May be I am missing MLEO (Multi level evolutionary optimization algorithm)

I think your project has the wrong schedule, if we can call it that: You are NOT required to utilize ANY metaheuristic algorithm, especially if it turns out that there are much better methodologies for the problem that you want to solve. I insist that you should *not* first start by providing a long list of algorithms, but to start by analysing the properties of your problem. Once you have done that you should design one or more methodologies to "attack" your problem. There are plenty of things to discuss as far as problem properties are concerned:

(a) Do you have an algebraic description of the problem - that is - explicit functions describing the objective function(s) and constraints?

(b) If not, how do you calculate the objective function, how long does it take, and is the calculation done with a good precision?

(c) What is the requirement on the result? Do you need a global optimal solution, or is it OK to get anything that is better than the best function value that you have at hand right now?

(d) Do you wish to solve many similar problems, with minor adjustments on problem paramaters, or just one problem instance?

You see the point? There is no purpose being narrow-minded; the best is to try several techniques; even if some fail, you will get some insight from the experience.

Michael Patriksson sir, I have a problem where I have no constraints only lower bound and upper bounds.I have a solution by using GRG method. Now I want to compare my result with well-known meta-heuristics method like PSO,Genetic algorithm,SA, Patternsearch , etc. And I think lots of researcher has the same type of problem how to implement and what should implement for their problem?

Then I need to ask why you are doing it: what particular insights do you hope to find from this exercise? Why do you want to run these experiments? With the information I have seen, I do no foresee that it will be publishable, for example. Perhaps if you have a very large test-bed of problems, including known hard ones with perhaps some having known solutions, you may provide a list of experiences regarding how accurate the results are from the various methods you have implemented. But what conclusions can you draw from it? Any test bed has its flaws; how will you construct yours, and how can you perform comparisons in a fair manner? These are not easy tasks! :-)

Darwin theory of evolution is great, but when dealing with an algorithm, there is still no silver bullet to globally optimize any problem (NP not= P?).

I would go with Genetics algorithms and a local optimization search.

I have developed the concept of the Integration of Variables method which central idea consist of making evolve any quantity of populations by any composition of operators. Based on this idea also developed various concrete evolutionary algorithms. In my researchgate profile you can find detailed information about and various applications related.

Dear Mashud,

I have some experiences in improving optimization algorithms like quantum invasive weed optimization algorithm (Article An Improved Quantum Evolutionary Algorithm Based on Invasive...

Article A Hybrid Neural Network – World Cup Optimization Algorithm f...

From my experiences two point are important to select a good optimization algorithm:

1) check that do your problems can be solved by the classic methods? if so, do not go to the meta-heuristics.

2) If your problem was NP-hard and can not solve by the classic methods:

after a lot of testifying, I found that there is no more differences among the evolutionary algorithms. of course in some case, one algorithm may have better performance or has high speed, but this prominence is not so bold.

3) Two cases that you should point to them (in evolutionary algorithms) are: Exploration and Exploitation.

Exploration parameter is for places that you have no information about the search space anymore and Exploitation is about that you have an approximate solution for your problem and this parameter in this case finds better solution.

These two parameters comprise the structure of all of the optimization algorithms. for example, in Genetic algorithm, Mutation is an exploration parameter and crossover is an exploitation parameter.

So based on your requirements, select an algorithm that has promonancy about your considered parameter.

good luck,