基于CNN-GRU-SE的自转着陆操纵策略研究

Research on Autorotation Landing Manipulation Strategy Based on CNN-GRU-SE

摘要:

针对直升机动力失效后的安全飞行问题，提出了一种基于CNN-GRU-SE预测直升机自转着陆过程中飞行员操纵策略的方法，该方法采用CNN-GRU-SE分阶段建立飞行员操纵模型。利用CNN强大的空间特征提取能力，结合GRU对时间序列数据的建模优势，以及SE模块的特征通道重校准功能，有效捕捉数据集中的时空相关性特征；再根据自转着陆特性，将过程划分为加速下降、稳定下降和减速着陆三个阶段，分别构建各阶段的操纵策略模型，进一步提升模型的针对性和准确性。通过模拟试飞数据进行模型训练与测试，结果表明该方法具有较高的预测精度，能够帮助飞行员在动力失效时，提供有效的操纵策略，从而提高安全着陆的可能。

Abstract:

To address the issue of safe flight following helicopter engine failure, a method based on a deep learning model is proposed to predict pilot manipulation strategies during autorotation landing. This approach employs the CNN-GRU-SE framework to establish phased pilot control models. By leveraging the powerful spatial feature extraction capabilities of CNN, the temporal sequence modeling strengths of GRU, and the channel recalibration function of the SE module, the method effectively captures spatiotemporal correlations in the dataset. The autorotation landing process is divided into three phases: rapid descent, stable descent, and deceleration landing, with specific manipulation strategy models constructed for each phase to enhance model specificity and accuracy. Model performance is validated using simulated flight test data, showing high prediction accuracy and the ability to provide effective manipulation strategies for pilots during engine failure, thereby increasing the likelihood of a safe landing.