基于改进的BP神经网络算法的控制系统在线辨识方法

 讨论了以计算机虚拟仪器为核心器件,搭建了动态测试与系统辨识硬件平台,使用Delphi语言编写辨识模块,实现对控制系统的在线辨识。在非线性系统辨识方面,针对BP神经网络算法中存在的收敛速度比较慢和辨识精度不高的问题,提出一种基于降低网络灵敏度的MBP神经网络辨识算法和一种基于小波分析的神经网络辨识算法,实现了对控制系统的状态进行预测估计。并以“防空武器半实物仿真系统”中的三轴稳定平台为对象,试验验证了算法的正确性。

An On-line Identification Method for Control System Based on Improved BP Neural Network Algorithm

The identification methods based on dynamic performance testing are discussed focusing on the features of control systems. In order to realize the online identification for the control system in a computer virtual instrument, a dynamic testing and identification system is designed, using Delphi language for several identification modules. The hardware platform of dynamic testing and system identification with the virtual instrument being used as its core component is built. In the nonlinear aspect, this paper studies the identification algorithm based on Back Propagation (BP) Neural Network algorithm, and proposes an identification algorithm named MBP Neural Network and a Wavelet Neural Network algorithm to reduce the sensitivity of the network. Finally, the real- time wavelet algorithm is simulated to verify the above conclusion. The MBP Neural Network algorithm and the Wavelet Neural Network algorithm are simulated on the three-axis platform of the partial simulating system for an anti-aircraft weapon.