What is it about?
A goal-oriented surrogate modeling framework is proposed to mitigate conventional surrogate models' reliance on extensive training datasets and predefined model architectures in high-dimensional nonlinear systems. The feature extraction methodology and subsequent validation through high-dimensional geometric parametrization demonstrate the efficacy of this approach, achieving enhanced predictive accuracy and improved generalization capabilities compared to baseline models.
Featured Image
Photo by Samuel’s Photos on Unsplash
Why is it important?
For the construction of high-dimensional models, input features either rely on knowledge-driven feature combination and construction methods (CST parameterization, etc.) or are reduced in dimension based on data-driven dimensionality reduction methods (proper orthogonal decomposition POD, etc.), which lacks coupling with design goals. This paper uses comparative learning methods to clarify the correlation between potential input features and target outputs, and proposes an output-driven input feature construction method. The constructed features have a significant enhancement effect in the construction of high-dimensional proxy models of numerical examples, airfoils, and wings. This work has discovered many interesting phenomena in the construction of model features, and we welcome your sharing and exchanges.
Perspectives
Data-driven surrogate models are widely used in aircraft design and optimization, significantly improving optimization efficiency and reducing the complexity of engineering systems. However, the application of high-dimensional surrogate models remains challenging due to efficiency and accuracy limitations. In this study, we propose a method for extracting hidden features to simplify high-dimensional problems and improve the accuracy and robustness of surrogate models. Specifically, we establish a goal-guided feature extraction (GFE) neural network. We then constrain the distance between hidden features based on the difference in target outputs. The proposed latent feature learning method significantly reduces the dimensionality and nonlinearity of the surrogate model, improving modeling accuracy and generalization. Numerical examples, airfoil, and wing modeling problems demonstrate that goal-guided feature extraction significantly improves modeling accuracy when the number of samples is insufficient for direct modeling, which will effectively expand the model's applicability. Comparisons of various data-driven models reveal that goal-guided feature extraction can also effectively reduce the error distribution of prediction examples, as well as the convergence and robustness differences caused by different data-driven surrogate models.
Xu Wang
Hong Kong Polytechnic University
Read the Original
This page is a summary of: Goal-Oriented Feature Extraction: A Novel Approach to Enhance Data-Driven Surrogate Models, AIAA Journal, September 2025, American Institute of Aeronautics and Astronautics (AIAA),
DOI: 10.2514/1.j065638.
You can read the full text:
Contributors
The following have contributed to this page
