Hello!
I have a question if someone can answer me please.
As for the majorioty of metaheuristic algorithm, a step called initialization is a necessary for the methods. There are some ways to complete the initialization such as random initializtion which require seting the upper and lower bound. Now, I wonder if the metaheuristic algorithm could be applied for calculating the unbounded problem.
I suggest you read the following articles:
https://ieeexplore.ieee.org/abstract/document/942532
Article Two-layer particle swarm optimization for unconstrained opti...
Why not look up one or both of the classic methods called Newton's method (with approximate line search, such as with Armijo's step length rule)? Or even quasi-Newton - easy to find a description on the web, and easy to implement in Matlab, or other tools for numerics.
It is almost considered a felony to replace Newton or variants thereof with heuristics, at least when you have access to second-order derivatives - typically you will more likely find local optima much faster with Newton's method.
I find it odd that so many ask this type of advice, when there are very good descriptions for free on the web.
Cheng Jiaming I really do recommend to read optimization literature, as well - and not just resort to heuristics. One can feed the other, too.
Programmatically, yes you can. However, optimization process it means to find the best solution within specific bounderies for variables ( you have variables length, width, weight, concentration, etc). For instance, you cant put the diameter of electrical wire 100m for transformer coil design. That doesnt make sense.
You should initialize your solution using bounded random solutions or give initial points to different heuristics available for the problem. Then, after generating initial solutions, you should update your solution according to position update method of the algorithm and in contrast to bounded problems, you do not need to check boundaries for the newly generated solutions.
You should not use heuristics at all, in fact - if you can avoid them.
I think heuristics can be used as a part of initialization. For instance, out of 50 population, position values of 5 members can be initiated using heuristics whereas the rest can be initialized randomly. Most probably heuristics generate better solution than random initialization. Therefore, it would be useful to use heuristics as a part of initialization to accelerate the convergence rate.
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory