Quantum support vector machines for aerodynamic classification

X.-J. Yuan, Z.-Q. Chen, Y.-D. Liu, Z. Xie, X.-M. Jin, Y.-Z. Liu, X. Wen, H. Tang · 2023 · Intelligent Computing（DOI 10.34133/icomputing.0057）

| 编号 QC061 | 主题 数据驱动方法 | 链接 arXiv:2208.07138 |

摘要

摘要保留原文（英文）

摘要取自 arXiv / 出版方元数据，未作翻译，以保证与原文一致。

Machine learning (ML) has flourished in fluid mechanics recently for its ability to extract informative features from data. It provides a powerful information processing framework that can be used for understanding, modeling, optimizing, and controlling fluid flows. ML is widely used in the flow reconstruction. The velocity fields can be faithfully predicted based on scattered pressure sensors and local measurements using Ensemble Kalman filter, artificial neural networks and dynamic mode decomposition with time delay [5–7]. Missing data in the flow measurement can also be reconstructed based on data fusion and deep generative adversarial model[8, 9]. The flow control parameter optimization and performance prediction can be achieved using support vector regression, model-free reinforcement learning and a deep Koopman embedding [10–12]. Classification is a classical problem for ML, which is also widely used in fluid dynamics to distinguish between various flow conditions and dynamic regimes. Support vector machines (SVMs) is one of the typical ML algorithms for classifying problem. It is a supervised learning method developed on the basis of statistical learning theory, embodying the structural risk minimization principle. They are gaining popularity due to the features such as simple structure, global optimality and strong robustness. Ling & Templeton [13] used SVMs to classify and predict regions of high uncertainty in the Reynolds stress tensor and compared the results with Adaboost decision trees and random forests. Wang et al [14] examined the feasibility of applying SVMs to high angle-of-attack unsteady aerodynamic modeling of an airfoil. Shi et al [15] applied SVMs to optimize spacecraft multidisciplinary design by using filter-based sequential radial basis function. In spite of the advantages above, SVMs are unsuited to implement for large-scale training samples due to enormous storage and calculation cost. Moreover, the general need for multi-class classification in real applications raise even more demand of the computation resources.

本站词条在引用该文献时，会以 wikilink（sources/QC061）链接到本页；右侧「反向链接」面板自动列出所有引用它的词条。