Hi, Khalid. As a matter of fact, though disappointing, both algorithms are of so-called meta-heuristics, of which the key point to get a successful application is to exploit the so-called domain-specific knowledge, for example, some relationship that can efficiently shrink the search space, or some heuristic rules that guide quick searches. So, it can be rudely concluded that no far better performance will be obtained when you hybrid them simply.

However, metahuristics incorporating domain-specific knowledge will never be a bad idea.

Good luck.

Hi,

I would recomend including domain specific knowledge within rules, that will reduce the search time signuficantly. Methods and their coupling depend on problem type, the nature of the search space. Usually the hybridisation includes coupling of metaheuristics with some of general mathematical programming approaches to speed up the search. Maybe to consider tunning of sesrch parameters on meta level rather than discussing a degree of hybridisation.

Hi,

In order to perform hybridization, you need to identify the strengths and weaknesses of both methods. Basically, you can use one then in sequence, just starting from the best solution found by the first one or run them in parallel and exchange their solution from time to time. But this may not be the best idea.

If one approach (A) is better at performing a global search and the other one (B) is better in the local exploration, then you can embed B within A. For instance Memetic Algorithms are Genetic Algorithms using local search or limited metaheuristic as special mutation operators. GRASP is also a good candidate for A and you can use any other metaheuristic as the local search: GRASP / Tabu Search, GRASP / Simulated Annealing, GRASP / VND, and so on.

As Tino mentioned, you can also think about a matheuristic, that is a combination of heuristic / metaheuristic and mathematical programming (exact) method.

Khalid,

I also concur with Duhamel in that you need to identify the strengths and weaknesses of both methods.

If you decide for a hybrid I would suggest taking an already established result such as the one from Chun, Lim, Jung and Yoon on "Optimal Design of Synchronous Motor with Parameter Correction Using Immune Algorithm," and replicate their results. After I establish the same result as theirs, I would use SA in their step 5 which is the production of antibodies that is based on mutation and crossover. Compare the results to see if they are good relative to their solution and then do the same comparison of both algorithms with your proposed dataset.

I think the modification of step 5 with SA would be a good possibility on which to benchmark the mutation and crossover against the annealing process.

Note: I have not searched to see if they have already done this modification to their algorithm.

Hope this helps

I think the problem itself is the key to decide which technique to apply to solve it. I agree that the point is the knowledge of the problem. Some questions you must answer before to use a technique.. it is better to begin trying to use a complete technique, so you can check optimality.

Given an initial algorithm (I), you must analyze which is its behaviour in terms of the diversification and intensification of the search space, could you change some parts of your design to improve it?, are the parameters used the best ones to have a good performance?... when you think that it will improve its performance if you include another technique, this second technique must be selected in order to help the algorithm (I) in terms of diversification and intensification. In general, hybridization means a more complex code, more parameters. It is important to have a tradeoff between complexity of the code and quality of the solutions. You can begin just including a local search procedure at the end, if your algorithm requires more intensification.

I think the blend to the extent at which hybridization is performed must be result driven. I have tested a hybrid of GA and SA in the evolution of neural network and while GA seems to be predominant on certain aspect of the evolution, SA helps to improve the overall parameter on the overall. Hence, the basic idea here is to ensure that the two algorithm that needs to be merged(creating a hybrid algorithm) need to co-exist well and need not to cancel out any good contribution brought forward by each of the individually. Moreover, to get a good hybrid of two algorithm, a study of the individual properties of each algorithm is vital. I sincerely hope i provide you an insight of my point of view and it proves to be vital in your quest.

Dear Mohd nor Akmal Khalid,

even if I'm not an expert in artificial immune systems. I think I can give you a little advice.

Often I hybridized NNs, GA, GPs, and FL, and always I got important improvements

in speed and/or goodness of solution et al.

The trick is to well study/understand the weaknesses and strengths of every component you want to use. After, the hybridization can give positive results if during the optimization phase you continuously check the status/quality/goodness of your temporary solution/solutions and from time to time choose the best component to move forward your hybrid optimization tool.

I hope this helps.