基于LSTM的辅助动力装置系统辨识与仿真  

System Identification and Simulation of Auxiliary Power Plant Based on LSTM

作　　者：杜维仲 王硕 Du Weizhong;Wang Shuo(Maintenance Engineering Department,Shenyang Branch of Shenzhen Airlines,Shenyang 110000,China)

机构地区：[1]深圳航空有限责任公司沈阳分公司维修工程部

出　　处：《计算机测量与控制》2020年第2期157-161,共5页Computer Measurement &Control

基　　金：国家自然科学基金重点项目(60832011);天津市科技攻关计划重点项目(06YFGZGX00700)

摘　　要：针对民航飞机辅助动力装置的复杂非线性时序特性,依据长短时记忆网络(LSTM)的长时序记忆、非线性关系表达能力,提出一种基于LSTM的辅助动力装置系统参数辨识模型;同时建模过程中考虑作用于辨识参数的多种影响因素,进一步提出基于LSTM的多变量时间序列预测参数辨识模型,解决了传统时序模型难以解决的多变量或者多输入问题;最后利用辅助动力装置试车样本,建立了APU启动阶段的参数辨识模型;仿真结果表明,发动机排气温度EGT温度均方根误差小于4℃,发动机转速百分比N均方根误差小于1,满足辅助动力装置仿真需求。Aiming at the complex nonlinear timing characteristics of civil aircraft auxiliary power plant,based on the long-term memory and nonlinear relationship expression of long and short time memory network(LSTM),a parameter identification model of auxiliary power plant system based on LSTM is proposed.At the same time,various influence factors of the identification parameters are considered in the modeling process,and the LSTM-based multivariate time series prediction parameter identification model is further proposed,which solves the multivariable or multi-input problem that is difficult to solve by the traditional time series model.Finally,using the auxiliary power device test sample,the parameter identification model of the APU startup phase is established.The simulation results show that the engine exhaust gas temperature root mean square error is less than 4℃,and the engine speed percentage N root mean square error is less than 1,which satisfies the auxiliary power plant simulation requirements.

关 键 词：LSTM 系统辨识 非线性系统 APU 深度学习 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]