| 引用本文: | 汪志刚,李爱军,王力浩,孙小锋.改进的输出误差法用于不稳定飞机参数辨识[J].哈尔滨工业大学学报,2022,54(12):65.DOI:10.11918/202103065 |
| WANG Zhigang,LI Aijun,WANG Lihao,SUN Xiaofeng.Improved output error method for parameter identification of unstable aircraft[J].Journal of Harbin Institute of Technology,2022,54(12):65.DOI:10.11918/202103065 |
|
| 摘要: |
| 为解决输出误差法在不稳定飞机参数辨识过程中的数值发散问题以及初值依赖问题，设计了一种结合神经网络、粒子群优化算法以及Levenberg-Marquardt算法的系统辨识方法。首先，为解决输出误差法的数值发散问题，以神经网络拟合待辨识系统的动力学特性。不同时刻的飞行试验数据用于训练神经网络，训练好的网络可以直接对下一时刻的运动状态进行预测，从而避免对不稳定运动方程的求解。其次，基于粒子群优化算法搜索Levenberg-Marquardt算法中的最佳阻尼因子，并以改进的LM算法替代输出误差法中的高斯-牛顿算法。接下来，改进的LM算法与训练好的神经网络结合得到了一种新的参数辨识算法。最后，基于不稳定飞机的闭环仿真飞行试验数据对提出的算法进行了验证。研究结果表明：与传统的最小二乘法和人工稳定的输出误差法的估计结果相比，所采用的算法具有更高的估计精度；同时，所提出的算法中可以随机选取待辨识参数的初值，克服了输出误差法对参数初值的依赖。本文的研究成果可以直接用于其他不稳定非线性动力学系统辨识领域，经过修改后还可以用于其他非线性优化领域。 |
| 关键词: 不稳定飞机 参数辨识 Levenberg-Marquardt算法 粒子群算法 神经网络 |
| DOI：10.11918/202103065 |
| 分类号:N945.14 |
| 文献标识码:A |
| 基金项目:航空科学基金(20180753005，201958053003) |
|
| Improved output error method for parameter identification of unstable aircraft |
|
Zhigang WANG1, Aijun LI1,2, Lihao WANG1, Xiaofeng SUN1
|
|
1.School of Automation, Northwestern Polytechnical University, Xi'an 710072, China;2.Shaanxi Province Key Laboratory of Flight Control and Simulation Technology (Northwestern Polytechnical University), Xi'an 710072, China
|
| Abstract: |
| Considering the problems of numerical divergence and initial value dependence of the output error method in parameter identification of unstable aircraft, a system identification method combining neural network, particle swarm optimization algorithm, and Levenberg-Marquardt (LM) algorithm was designed. First, in order to solve the numerical divergence problem of the output error method, the neural network was utilized to approximate the dynamic characteristics of the system to be identified. The flight test data at different moments were used to train the neural network. The trained network could directly predict the motion state at the next moment, so as to avoid solving the unstable motion equation. Then, the particle swarm optimization algorithm was adopted to search the best damping factor in LM algorithm, and the improved LM algorithm was used to replace the Gauss-Newton algorithm in the output error method. Next, the improved LM algorithm was combined with the trained neural network to form a new parameter identification algorithm. Finally, the proposed algorithm was verified based on the closed-loop simulation flight test data of unstable aircraft. Research results show that compared with the estimation results of the traditional least square method and output error method with artificial stabilization, the proposed algorithm had higher estimation accuracy, and it could randomly select the initial value of the parameters to be identified, which overcomes the dependence of the output error method on the initial value of the parameters. The research results of this paper can be directly used in the identification of other unstable nonlinear dynamic systems, as well as other nonlinear optimization fields after modification. |
| Key words: unstable aircraft parameter identification Levenberg-Marquardt algorithm particle swarm optimization neural network |
