A Review of Surrogate Assisted Multiobjective Evolutionary Algorithms
Table 1
Advantages and disadvantages of each class in the taxonomy.
Class
Advantages
Disadvantages
DFR
No evolution control
(i) Computationally efficient (ii) Good behavior in low-dimensional problems
(i) Requires an accurate surrogate model (ii) It can converge to a false optimum
Fixed evolution control
(i) It is capable of solving high-dimensional problems (ii) The surrogate model adapts during the optimization process
(i) It is necessary to define the parameter to alternate between the surrogate model and the real objective function
Adaptive evolution control
(i) Does not require defining the parameter to alternate between the surrogate model and the real objective function
(i) The automatic alternation is not easy to define
IFR
(i) Usually uses a local search phase to optimize the surrogate model (ii) The metamodel is used for exploitation purposes (iii) Avoid convergence to a false optimum
