A multi-objective variable-fidelity optimization method for genetic algorithms

A survey on handling computationally expensive multiobjective optimization problems with evolutionary algorithms

“…A multiobjective variable-fidelity optimization (VFO) algorithm was proposed in [151], where Kriging was used as a metamodel with NSGA-II. Initially, a simplified or approximated problem (having low fidelity functions) is used to replace the original MOP in stage 1.…”

“…The number of papers which used benchmark and real-world problems is also mentioned in Figure 3. Seven algorithms [84,104,107,110,127,131,148] were tested on benchmark problems, seven algorithms [6,48,52,59,75,94,128] on real-world problems and seven algorithms [14,81,83,85,86,90,151,18] on both. The efficiency of the different algorithms in terms of computation time or number of function evaluations reduced is very important, especially in the case of real-world problems.…”

“…As shown in Figure 3, single metamodel is used more than multiple metamodels and ensemble of metamodels. In algorithms using single metamodel, Kriging [59,83,81,86,110,128,148,151] has been used more than other metamodels such as neural networks [14,76,94,104,131], support vector regression [48,107] and polynomial regression [85].…”

“…In [151], Kriging was used as a metamodel and the authors mentioned that within different metamodels studies to date, perhaps the most common type of metamodel used is the Kriging model. In [14], radial basis functions (RBFs) were used and the authors mentioned that RBF was a kind of approximation having a very good approximation accuracy.…”

“…As mentioned in Section 2, using an adaptive evolution control strategy depends on the accuracy of the metamodel used and using a fixed evolution control strategy implies that either the metamodel is accurate enough for approximation or the metamodel accuracy is not important or not checked. There are only five algorithms [86,94,95,127,151], where adaptive evolution control was used. For instance in [86], after using Kriging models, uncertainty of the approximated values was compared with a predefined value and if that uncertainty was acceptable, then only Kriging models were further used otherwise original function were used.…”