Considering:
- test problems
- quality indicators used in the evaluation
Global optimization, which combines genetic algorithms, differential evolution, particle swarm optimization, and adaptive simulated annealing. It is a robust method, which combines single, and multi-objective optimization by adding objective functions.
You can go through the paper " Discrete Time-Cost Tradeoff with a Novel
Hybrid Meta-Heuristic" by
Kamal Srivastava, Sanjay Srivastava, Bhupendra. K. Pathak,
and Kalyanmoy Deb
This method hybridizes SA with GA for MOOP
My experience - models seem to express different levels of search behavior for each different problem and representation of the data. SA has been shown to be very good in some forest planning problems, for example, but other models (TS and TA) can produce better solutions when the parameterization and formulation of the metaheuristic is considered.
Consider Simulated Annealing is fairly efficient when undergoing problems under loads of stress functions is when the universe of solutions available is dynamic, ie cambainte; However compared with other heuristics in standard conditions or under stress SA is not as efficient. With regard to multi-objective algorithms, I believe that these cases are not suitable for data traffic which is my expertiz; and cutting conditions are based on a single target. However, depending on the strategy followed, we experimented using weights (pesos per target) and excellent results.
There is nothing called Most competitive multi-objective SA. The cooling schedule is the necessary condition for convergence , not the sufficient.
Cooling schedule differs from problem to problem and mostly to be ascertained by trial methods. Hajek's proof is impractical from computational point of view.
The question is which is the most efficient MOSA algorithm (i.e. Multiobjective Simulated Annealing or SA). Bettinger says TS and TA are good options; however, these algorithms are not SA algorithms. The test that an algorithm is better than another one is hard to do. How many instances of the entire set of Multiobjective problems are needed to test this hypothesis? Probably there are some multiobjective problems where even exact methods are the best, while in other cases another algorithm different to TS or TA could be the best; we'll never know.
Until now I don't know a publication showing which is the best of the MOSAs algorithms.
Simulated Annealing is good but the cooling schedule needed to be properly monitored because of the fact that it determines the quality of the final output. Most of the time, since its problem-dependent, it must be properly studied for accuracy
- Badges
- Science topic