Table 4 Surrogate-assisted algorithms used to benchmark the proposed MO procedure with variable-fidelity machine-learning.
Algorithm | General information | Surrogate model | Characterization |
|---|---|---|---|
1 | One-shot surrogate-assisted MO procedure | Kriging interpolation (Gaussian correlation functions and first-order polynomial as a trend function84) | Surrogate constructed using NS data samples, then optimized using MOEA; Selected non-dominated samples evaluated with EM analysis to form the final outcome of the algorithm |
2 | Neural network (setup: multi- layer perceptron, two hidden layers with ten neurons each; training: Levenberg-Marquardt algorithm) | ||
3 | Machine learning algorithm with ANN surrogates | ANN surrogates: Initial sampling and surrogate model setup the same as discussed in section “Initial sampling. ANN regression models”; Infill point generation as discussed in sections “Multi-objective evolutionary algorithm” and “Infill point generation” (surrogate optimization using MOEA); EM-simulation dataset updates by adding all infill points to the existing dataset | Search process executed using a single (high-fidelity) EM simulation model |
