基于Transformer的飞机状态预测  

Aircraft state prediction based on Transformer

作　　者：王经纬 高艳鹍 宋澣兴 刘一非 WANG Jing-wei;GAO Yan-kun;SONG Huan-xing;LIU Yi-fei(China Aviation System Engineering Research Institute,Chinese Aeronautical Establishment,Beijing 100029,China;Institute 706,Second Academy of China Aerospace Science and Industry Corporation,Beijing 100854,China;School of Computer Science,Beijing University of Posts and Telecommunications,Beijing 100876,China;School of Computer Science,Beijing Technology and Business University,Beijing 100048,China)

机构地区：[1]中国航空研究院中国航空系统工程研究所,北京100029 [2]中国航天科工集团第二研究院七〇六所,北京100854 [3]北京邮电大学计算机学院,北京100876 [4]北京工商大学计算机学院,北京100048

出　　处：《计算机工程与设计》2024年第2期477-483,共7页Computer Engineering and Design

摘　　要：在非定常气动力下,为防止飞机进入危险状态,通过建模进行状态预测,是保障飞行安全的重要手段,传统方法建模过程复杂、工程化难度大且普适性不强。为更好解决大迎角下飞行状态预测,使用基于深度学习的时序序列预测方法,推测飞机的飞行状态,达到最大限度发掘飞机性能、保障飞行安全的目的。提出一种多任务Transformer模型,同时完成飞行状态参数回归和飞行状态分类。实验结果表明,相比于同类模型,该模型的预测性能有大幅提升。To prevent aircraft from entering dangerous state under unsteady aerodynamics,the state prediction by modeling is important means to ensure flight safety.Traditional methods have complex modeling processes,high engineering difficulty,and weak universality.To better solve the prediction of flight status at high angle of attack,a time series prediction method based on deep learning was used to infer the flight status of the aircraft,thereby achieving the goal of maximizing aircraft performance and ensuring the flight safety.A multi task Transformer model was proposed to simultaneously perform flight state parameter regression and flight state classification.Through experiments,compared to similar models,the predictive performance of this model is significantly improved.

关 键 词：多任务 深度学习 时序预测 状态分类 气动力建模 大迎角 非定常气动力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]