The differences are quite large. As you say, one of them is based on Pareto, and the other is not. MOO is a huge thing. VEGA works by breaking a population into disjoint sets and forcing different sets to evolve towards different goals. There are few mechanisms there to help combine them into a meaningful presentation of the Pareto set, but it is basically just a combination of solutions for different purposes. The choice is made by choosing the solutions that are best suited for their individually defined objective functions.
NSGA and other Pareto based methods are completely different. They make a choice, not based on any specific choice of purpose, but on the properties of decisions in comparison with each other. Each such algorithm makes slightly different options in the way they perform these comparisons, and NSGA-II (you should definitely use the second version of the algorithm) does this by non-dominant sorting. Basically you will find all non-dominant solutions and name them set # 1. Then you will find all solutions that will not dominate if you delete the elements of set # 1 - they will become installed # 2. You keep going until you all decisions are taken into account, and the result will be similar to peeling onion layers. The selection procedure is that you always select members of the lower classes (set # 1, then # 2, etc.). If you cannot take all the elements of a certain level, you break the connection by choosing solutions at this level that are further from the others, the idea is that if you cannot take all of them, you should at least try not to choose the ones you take from one small cluster.
In general, you should look at Pareto based methods. They have been a proven choice for at least 10-15 years. In particular, you should focus on elite Pareto-based techniques such as NSGA-II, SPEA2, epsilon-MOEA and a few more recent rivals.
deong source share