付军泉, 史志伟, 陈坤, 朱佳晨, 陈杰, 董益章. 基于EKF的实时循环神经网络在非定常气动力建模中的应用[J]. 空气动力学学报, 2018, 36(4): 658-663. DOI: 10.7638/kqdlxxb-2016.0131
引用本文: 付军泉, 史志伟, 陈坤, 朱佳晨, 陈杰, 董益章. 基于EKF的实时循环神经网络在非定常气动力建模中的应用[J]. 空气动力学学报, 2018, 36(4): 658-663. DOI: 10.7638/kqdlxxb-2016.0131
FU Junquan, SHI Zhiwei, CHEN Kun, ZHU Jiachen, CHEN Jie, DONG Yizhang. Applications of real-time recurrent neural network based on extended Kalman filter in unsteady aerodynamics modeling[J]. ACTA AERODYNAMICA SINICA, 2018, 36(4): 658-663. DOI: 10.7638/kqdlxxb-2016.0131
Citation: FU Junquan, SHI Zhiwei, CHEN Kun, ZHU Jiachen, CHEN Jie, DONG Yizhang. Applications of real-time recurrent neural network based on extended Kalman filter in unsteady aerodynamics modeling[J]. ACTA AERODYNAMICA SINICA, 2018, 36(4): 658-663. DOI: 10.7638/kqdlxxb-2016.0131

基于EKF的实时循环神经网络在非定常气动力建模中的应用

Applications of real-time recurrent neural network based on extended Kalman filter in unsteady aerodynamics modeling

  • 摘要: 结合EKF算法(扩展卡尔曼滤波)和RTRL算法(实时递归学习算法)的特点,提出一种基于EKF的实时递归学习算法(EKF-RTRL),运用到循环神经网络中(RNN)。应用该神经网络对某飞机大迎角大振幅单自由度偏航、滚转以及偏航滚转耦合运动的非定常气动力进行建模。结果表明,基于EKF的实时循环神经网络计算精度高,收敛快,辨识结果与实验结果符合较好,验证了本算法的有效性。

     

    Abstract: According to the characteristics of extended Kalman filter (EKF) and real-time recurrent learning (RTRL) algorithm, the EKF-RTRL algorithm is applied to recurrent neural network (RNN). Based on the large amplitude harmonic oscillation and ramp motion experimental data of an aircraft model with yawing, rolling, and yaw-roll motion, the application of the RNN in modeling unsteady aerodynamics are studied. The results show that, this recurrent neural network has advantages of high computing precision and fast convergence feature. A good agreement is observed between the simulation results and testing results. It indicates that the neural network modeling method is efficient.

     

/

返回文章
返回