【作者】 崔杰；

【导师】 宋申民；

【作者基本信息】 哈尔滨工业大学 ， 导航、制导与控制， 2006， 硕士

【摘要】 倾斜转弯(BTT)技术用于导弹控制越来越受到重视。该技术控制导弹捕捉目标的过程中,能实现导弹绕纵轴转动,使其所要求的理想法向过载矢量总是落在导弹的最大升力面上。因此,与STT导弹相比,BTT导弹具有更高的机动作战能力。但是另一方面,相对STT导弹,BTT导弹在俯仰、偏航方向存在强烈的耦合项。因此,BTT导弹的整个模型呈现出高度非线性。传统的三通道独立设计自动驾驶仪的方法不适合于BTT导弹控制方式。这给BTT导弹自动驾驶仪的设计带来了很大的挑战。本文利用自适应神经网络并结合H∞控制理论来设计BTT导弹自动驾驶仪,该设计方法克服了传统自动驾驶仪在控制导弹飞行中需逐段切换增益的不足,控制BTT导弹能够满足跟踪性能指标。本文研究的主要工作有以下几个方面:首先是研究了BTT导弹数学模型。根据BTT导弹的数学模型是一类具有强耦合、非线性、时变的非最小相位系统的特点,利用输入—输出反馈线性化来处理BTT导弹的数学模型。同时针对该方法不能直接应用到非最小相位系统的问题,对BTT导弹的输出进行了重新定义。然后,针对输入—输出反馈线性化得到的数学模型中的非线性项,本文利用神经网络来对该部分进行逼近。在设计过程中,神经网络采用了脊型高斯神经网络。该网络和一般的神经网络相比,具有两方面优点:局部逼近能力好和在线调节参数相对较少。从其结构上来说,实质上是高斯径向基函数的一种扩展。理论上,脊型高斯神经网络能够以任意精度逼近非线性系统。但是,由于实际逼近过程中存在着逼近误差,为了保证设计出的控制器的性能,需要利用H∞控制理论来进行控制器的设计。从而使设计的控制系统具有良好的鲁棒稳定性和良好的跟踪性能。最后,根据建立的BTT导弹非线性时变控制系统数学模型,进行了仿真研究。对BTT导弹控制系统的性能指标进行评价以及在导弹控制系统跟踪导引指令的性能和鲁棒性能等。仿真结果表明,本文所设计基于H∞理论和自适应神经网络所设计的自动驾驶仪正确可行,能较好的满足设计目标要求。更多还原

【Abstract】 Bank-to-Turn (BTT) missiles have arosen increasing attention since its coming into being. This is due to the fact that the normal over-loading vector of this missile can always be located at the missile’s maximum lifting surface. Hence BTT missiles feature superior maneurability to Skid-to-Turn (STT). However, BTT surfers strong coupling in the pitch and yaw motions, which exhibits highly nonlinearity in the control system. Thus the traditional design methodology of STT missiles’autopilots, which treats the three loops independently, proves to be non-sufficient to BTT ones. This brings great challenge to the designing of BTT missile’s autopilot.To solve this problem, this thesis integrates adaptive neural network and H-infinity control theory to design BTT missile’s autopilot. This method has overcome the shortcomings of the traditional design methods. It needs no piecewise switching to control the BTT missile to realize the performance specification. The main contents of this thesis are as following:Firstly, the mathematical model of BTT missiles is established. Since this model is a highly-coupled, non-linear, time-varying and non-minimum phase system, input-output linearization is used to do some preprocessing work. Meanwhile, since that methodology can not be applied to non-minimum phase system, BTT missiles’outputs are re-formulated.Then a neural network is employed to approximate the non-linear component of the input-output linearized system. This network is a ridge Gaussian one, which is preferred because of its good approximating capability and less online tuning weights. As for its architecture, it is simply an extention of Gaussian radian basis function. Thus theoretically this network can approximate any nonlinear system with any precision.However, since the approximating error is almost always present, H-infinity theory is refered here to design a controller to guarantee the performance of the whole control system.Furthermore, some simulations are conducted with the mathematical models aforementioned. These simulations aim to reveal the BTT control system’s更多还原