Abstract:The optimization of the engineering system is becoming more and more significant as time progresses, and a large number of metamodel-based multi-objective optimization methods have been proposed during the past decades. For a metamodel-based multi-objective optimization method, the accuracy of the solution is greatly influenced by the prediction accuracy of the metamodels. This study developed a novel metamodel-based multi-objective optimization method using an adaptive multi-regional ensemble of metamodels (AMEM) to get an optimization method with higher accuracy and efficiency. Two strategies were employed in this new metamodel-based multi-objective optimization method. One strategy is that several stand-alone metamodels, i.e., polynomial regression (PR), radial basis function (RBF), and Kriging (KRG) were combined into an ensemble of metamodels (EM) using the weight sum approach. The other strategy is that new design points were dynamically added into various local regions where the Pareto optimal designs are located, and each basis model and its weight factors were regenerated simultaneously, which adaptively improved the accuracy of the constructed adaptive multi-regional ensemble of metamodels. Besides, two novel approaches for selecting new design points from Pareto optimal set were proposed in this study. The performance of this novel metamodel-based multi-objective optimization method was evaluated using twelve mathematical functions and two practical engineering optimization problems. The twelve mathematical functions were typically selected from previous studies, and the two practical engineering optimization problems are crashworthiness optimization of triply periodic minimal surface (TPMS)-filled tube (one design variable) and buffer performance optimization of airbags used in the re-entry capsule (two design variables). To compare the advantages of this method, the optimization of these problems was also implemented by common metamodel-based methods. The results showed that the adaptive multi-regional ensemble of metamodels-based multi-objective optimization method is more accurate than the common methods in both mathematical functions and engineering optimization problems. In addition, the adaptive multi-regional ensemble of metamodels-based multi-objective optimization method has better efficiency than the common method, especially in practical engineering optimization problems.

A New Adaptive Multi-Fidelity Metamodel Method Using Meta-Learning and Bayesian Deep Learning

An Active Learning Multi-Fidelity Metamodeling Method Based on the Bootstrap Estimator

Metamodeling for High Dimensional Design Problems by Multi-Fidelity Simulations

A Sequential Multi-Fidelity Metamodeling Approach for Data Regression

Adaptive Radial-Basis-Function-Based Multifidelity Metamodeling for Expensive Black-Box Problems

Hyperparameter-optimized Multi-fidelity Deep Neural Network model associated with Subset Simulation for structural reliability analysis

Multi-fidelity deep neural network surrogate model for aerodynamic shape optimization

A Latent Variable Approach for Non-Hierarchical Multi-Fidelity Adaptive Sampling

A Point Cloud Deep Neural Network Metamodel Method for Aerodynamic Prediction

A multi-fidelity active learning method for global design optimization problems with noisy evaluations

Deep Multi-Fidelity Active Learning of High-dimensional Outputs

Adaptive multi-fidelity sparse polynomial chaos-Kriging metamodeling for global approximation of aerodynamic data

A Novel Metamodel-Based Multi-Objective Optimization Method Using Adaptive Multi-Regional Ensemble of Metamodels

An Active Learning Variable-Fidelity Metamodelling Approach Based on Ensemble of Metamodels and Objective-Oriented Sequential Sampling

A Multi-Fidelity Bayesian Optimization Approach Based on the Expected Further Improvement

Disentangled Multi-Fidelity Deep Bayesian Active Learning

Multi-Fidelity Bayesian Neural Network for Uncertainty Quantification in Transonic Aerodynamic Loads

A new multi-fidelity surrogate modelling method for engineering design based on neural network and transfer learning

Non-myopic multipoint multifidelity Bayesian framework for multidisciplinary design

A deep learning-based multi-fidelity optimization method for the design of acoustic metasurface