Comparison of EGO and sEGO optimization algorithm based on Kriging for noisy function

Resumo

In many engineering optimization problems the number of function evaluations is severely limited by time
or computational cost. In addition, the representation of randomness due to noise and uncertainties in the model
is essential. One strategy adopted for these cases is solve the problem through response surfaces, or meta-models,
especially Kriging model. A traditional Kriging-based algortihm optimization is the Global Efficient Optimization
(EGO) method. An most recent algorithm for stochastic problems was sEGO in which it introduces a parcel that
reflects the intrinsic noise of the stochastic function in your framework. In this paper these optimization algorithms
will be approached through some examples for demonstrate the importance of the variance quantifying approach
in the optimization process through Kriging meta-model, highlighting the influence of the noise amplitude in the
choice of the optimization strategy. The conclusions obtained may serve as a guideline for choose the best approach
for each type of optimization problem.