基于PSO算法的WNN大包线增益调参控制律设计  

Design of a Large Envelope Wavelet Neural Network Gain Scheduling Flight Control Law Based on PSO Algorithm

作　　者：孙逊[1] 章卫国[1] 尹伟[1] 李爱军[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《测控技术》2007年第10期34-36,共3页Measurement & Control Technology

摘　　要：提出了一种基于粒子群优化算法的小波神经网络大包线调参控制律设计方法。该方法用小波函数代替了Sigmoid函数作为激活函数。由于结合了小波变换良好的高频域时间精度、低频域频率精度的性质和神经网络的自学习功能,因而具有较强逼近非线性函数的能力。为了克服局部极小值问题并进一步提高对非线性函数逼近能力,利用粒子群优化算法对小波神经网络进行参数训练,并利用该网络实现了大包线增益调参。飞行仿真结果表明,所设计的小波神经网络增益调参控制器具有优良的控制性能,不仅能够保证平衡状态下的控制效果,而且在未训练的平衡状态下依然具有良好的控制性能,并且在存在20%的建模误差时,最大超调量仅为6 m,仅是使用常规增益调参方法的18%。A design method of a large envelope wavelet neural network gain scheduling flight control law based on PSO algorithm is presented. In this method, Sigmoid function is replaced by wavelet function as the excitation function. So the good learning character and good localization character are merged and the high nonlinear function approximation capability can be achieved. In order to o- vercome the local minimum and improve the nonlinear function approximation capability, PSO algorithm is used to train the parameters of the wavelet neural network. This method is used in design of the large envelope gain scheduling flight control law. The flight simulation results show that good control capability can be achieved in untrained flight conditions and the system is still stable when modeling error is 20%. In case of 20% modeling error, the maximum overshoot is 6 m and it is only 18% of the maximum overshoot using traditional method.

关 键 词：粒子群优化 小波神经网络 增益调参 大包线飞行控制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] V448[自动化与计算机技术—控制科学与工程]