基于CapsNet的飞控参数预测算法  

FLIGHT CONTROL PARAMETER PREDICTION ALGORITHM BASED ON CAPSNET

作　　者：张鹏 刘颖欣 段照斌 王力[4] Zhang Peng;Liu Yingxin;Duan Zhaobin;Wang Li(College of Airworthiness,Civil Aviation University of China,Tianjin 300300,China;College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;Engineering Training Center,Civil Aviation University of China,Tianjin 300300,China;College of Vocational and Technical,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学适航学院,天津300300 [2]中国民航大学电子信息与自动化学院,天津300300 [3]中国民航大学工程技术训练中心,天津300300 [4]中国民航大学职业技术学院,天津300300

出　　处：《计算机应用与软件》2021年第12期36-41,共6页Computer Applications and Software

基　　金：国家自然科学基金委员会与中国民用航空局联合资助项目(U1733119)。

摘　　要：针对飞行控制系统(Flight Control System,FCS)一直以来难以进行故障预测的问题,提出一种基于胶囊网络(Capsule Network,CapsNet)的飞控参数预测算法。通过将飞控系统的相关多个参数融合输入到模型中来实时预测单个参数在飞机飞行中的变化,从而可以在其发生故障之前及时排除。在基于Keras框架上进行的实验表明:在四种模型性能评估指标上,CapsNet的方法比传统的卷积神经网络(Convolutional Neural Network,CNN)、长短时记忆网络(Long Short-Term Memory,LSTM)在单步以及多步预测上误差平均降低37.1%、8.1%,可以为飞控系统故障预测提供重大参考。Aiming at the problem that the flight control system(FCS)is difficult to predict faults,we propose a flight control parameter prediction algorithm based on Capsule Network(CapsNet).The changes of the individual parameters in the flight of the aircraft could be predicted in real time by incorporating the relevant parameters of the flight control system into the model,so that the faults could be removed in time before they occurred.The experiments based on the Keras framework show that in terms of four model performance evaluation indexes,CapsNet method reduces the average errors of single-step and multi-step prediction by 37.1%and 8.1%compared with traditional convolutional neural network(CNN)and long short-term memory(LSTM).It can provide a major reference for fault prediction of flight control system.

关 键 词：FCS CapsNet 参数预测 CNN LSTM 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]