There's no reason it can't be. And there are many applications-it suffices to search.

Dear Sumit,

You can check this articles below:

Article Application of the Genetic Algorithm to Real-Time Active Noi...

Article Genetic Algorithms Applied To Real Time Multiobjective Optim...

Hope this will help you

In most of the cases genetic algorithms are time consuming and given solutions are sub-optimal.

Suboptimal compared to what? And ``time consuming'' doesn't mean anything, either. Neither imply that genetic algorithms can't be used for real time applications.

Dear Sumit, It very much depends on the problem timescales and also what quality of solution is acceptable. First of all, you must be sure that the problem cannot be solved adequately using classical optimisers before attempting with the Evolutionary methods; there is no point in using a GA when either a closed-form solution is available, or the problem can be solved using classical gradient methods. If classical methods all fail, then stochastic optimsers may be practical (and possibly the only choice). The advantage of an EA over methods such as simulated annealing are that in an EA, most of the solution quality is established in the early generations where population convergence is rapid. In simulated annealing, the early phases of the algorithm are a wide exploration and the best results tend to occur towards the end of the run. Therefore if time is short, stopping an EA early will often produce something that is quite useful (but not fully optimal of course at that time). Real time problems, such as dynamic scheduling, will often be making small changes to a good existing solution, and therefore often a full re-optimisation is not needed, saving time too. Often it is best to try to reformulate the problem so that the work to be done by the EA is minimised. For example using the EA to find a promising region of the search space that can then can have bounds and constraints applied locally that enables a fast gradient-based method to be applied can work very well. Any time a repeated random start gradient method can be used to solve a multi-modal problem, an EA+gradient is almost always superior anyway. Evan