Scholarlink

The variations of unsteady aerodynamic loads during aircraft maneuvers are complex, which necessitates high-fidelity modeling to ensure operational safety in extreme flight conditions. In response to the high-precision prediction requirements of unsteady aerodynamic loads during rapid maneuvering of aircraft, in this study, a modeling method based on Long Short-Term Memory (LSTM) networks was proposed for unsteady aerodynamic loads. A systematic sensitivity analysis of critical hyperparameters – including loss functions, activation functions, and temporal window length – was conducted to identify optimal architectural settings. Validation against large-amplitude forced oscillation test data of the same origin demonstrated exceptional predictive accuracy. The model maintained robust performance in maneuvering motion trajectory datasets, verifying its strong generalization ability for predicting unseen maneuver patterns. The LSTM networks effectively resolved nonlinear hysteresis effects through its gated temporal dependency modeling and captured transient flow separation characteristics via implicit frequency-domain feature extraction, establishing its capability to capture nonlinear hysteresis effects and transient flow separation characteristics in changes of unsteady aerodynamic loads.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hyperparameter Optimization and Engineering Validation of LSTM-Based Unsteady Aerodynamic Loads Modeling

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jianing Wang,
  • Yanling Wang,
  • Yanjie Shen,
  • Chen Bu

摘要

The variations of unsteady aerodynamic loads during aircraft maneuvers are complex, which necessitates high-fidelity modeling to ensure operational safety in extreme flight conditions. In response to the high-precision prediction requirements of unsteady aerodynamic loads during rapid maneuvering of aircraft, in this study, a modeling method based on Long Short-Term Memory (LSTM) networks was proposed for unsteady aerodynamic loads. A systematic sensitivity analysis of critical hyperparameters – including loss functions, activation functions, and temporal window length – was conducted to identify optimal architectural settings. Validation against large-amplitude forced oscillation test data of the same origin demonstrated exceptional predictive accuracy. The model maintained robust performance in maneuvering motion trajectory datasets, verifying its strong generalization ability for predicting unseen maneuver patterns. The LSTM networks effectively resolved nonlinear hysteresis effects through its gated temporal dependency modeling and captured transient flow separation characteristics via implicit frequency-domain feature extraction, establishing its capability to capture nonlinear hysteresis effects and transient flow separation characteristics in changes of unsteady aerodynamic loads.