自组织模糊神经网络的气动参数辨识算法  

A Self-Organizing Fuzzy Neural Network Algorithm for Aerodynamic Parameter Identification

作　　者：梁学明[1,2] 梁晓庚[3] 杨士元[1] 余熙武[2] 

机构地区：[1]清华大学自动化系,北京100084 [2]北京航空工程技术研究中心,北京100076 [3]洛阳光电技术发展中心,河南洛阳471009

出　　处：《电光与控制》2010年第2期1-4,9,共5页Electronics Optics & Control

基　　金：航空科学基金资助(20070112001)

摘　　要：导弹气动参数一般通过理论计算或风洞试验数据来获取,由于受诸多因素限制而难以获得精确值。为了提高导弹动力学模型的精确度,利用神经网络能够逼近任意非线性函数的特点,对各实际工作点的气动参数进行辨识。采用反向传播神经网络的结构模型,通过加入模糊推理算法进行神经元权值修正的聚类训练,并动态地调节隐层节点数目,由此提出了一种能够精确跟踪非线性函数的网络模型。网络中若不存在能代表某些输入的聚类中心的神经元时,增加隐层节点数即相当于增加相应的模糊规则,它体现了该网络的自组织特点。仿真结果验证了模型对气动参数辨识的有效性,该方法对于自动驾驶仪气动参数的修正具有良好应用前景。Missile aerodynamic parameters are generally obtained from the theoretical calculation or wind tunnel experiment data, it is difficult to get the exact value because of the limitation of a great deal of factors. In order to improve the accuracy of the dynamics model of missile, the characteristic of Neural Network（NN） that can approximate arbitrary nonlinear function is used in the identification of aerodynamic parameters to each actual operating point. Structure of counter propagation network is employed, the neuron values are adjusted for the cluster training with the help of fuzzy inference algorithm, and the number of hidden layer nodes can be modified. Thus a fuzzy network model which can precisely track a nonlinear function is proposed. If there is no neurons in network which can be on behalf of the cluster center for some input, the number of nodes in hidden layer is increased, which is equivalent to an increase of the corresponding fuzzy rules. It reflects the characteristics of self-organizing in this fuzzy network. The simulation results verify the effectiveness of this model for aerodynamic parameter identification. The method has a good application prospects for autopilot aerodynamic parameter amendment.

关 键 词：导弹自动驾驶仪 模糊推理 反向传播神经网络 气动参数 参数辨识 

分 类 号：V271.4[航空宇航科学与技术—飞行器设计] TP301[自动化与计算机技术—计算机系统结构]