Data-Driven Evolutionary Algorithm With Perturbation-Based Ensemble Surrogates

Data-driven evolutionary algorithms (DDEAs) aim to utilize data and surrogates drive optimization, which is useful efficient when the objective function of optimization problem expensive or difficult access. However, performance DDEAs relies on their surrogate quality often deteriorates if amount available decreases. To solve these problems, this article proposes a new DDEA framework with perturbation-based ensemble (DDEA-PES), contain two mechanisms. The first diverse generation method that can generate through performing perturbations data. second selective selects some prebuilt form final model. By combining mechanisms, proposed DDEA-PES has three advantages, including larger quantity, better utilization, higher accuracy. validate effectiveness framework, provides both theoretical experimental analyses. For comparisons, specific algorithm developed as an instance by adopting genetic optimizer radial basis neural networks base models. results widely used benchmarks aerodynamic airfoil design real-world show outperforms state-of-the-art DDEAs. Moreover, compared traditional nondata-driven methods, only requires about 2% computational budgets produce competitive results.