I have applied NSGA-III using the pymoo library on my problem having 6 Objective functions.
The results are not so impressive. However, I applied epsilon constraint algorithm and it gave me best solutions.
What possible problem I can expect of not having good results from NSGA-III?
(1) If you have more than 3 objectives, then you cannot verify the obtained Pareto front via 2D or 3D plots. Even the true Pareto front for 6 objectives might look strange in 2D or 3D plots.
(2) Maybe your problem is beyond the capabilities of this algorithm. There is a subset of multi-objective problems called many-objective optimization, where the number of objectives are much larger than conventional 2 or 3. Maybe many-objective algorithms are better for you.
(3) Is your implementation correct? You should first validate your NSGA-III code on benchmark functions whose exact Pareto front are known. You should start with a verified source code that you are sure that it is tested and is implemented correctly.
(4) Are your objective functions conflicting? If your objective function are not conflicting, then you might get non-conventional plots for obtained Pareto front.
(5) Multi-objective problems need much more computing budget than single-objective optimization; i.e., you need to increase population size, max generations, or other parameters to give the algorithm more resources to search for solution.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python