【作者】 孙平；

【导师】 井元伟；

【作者基本信息】 东北大学 ， 导航、制导与控制， 2005， 硕士

【摘要】 传统的基于模型的控制方式,是根据被控对象的数学模型及控制性能指标来设计控制器。而对于非线性系统来说,被控对象复杂且多具有不确定性,因此建模和控制用传统的方法难以实现。人工神经网络在不同程度上模仿人脑神经系统的信息处理、存储和检索功能,可以逼近任意非线性函数,具有并行性、自学习自组织性、容错性、联想性等诸多优点。因此可以在非线性控制方面解决许多问题。目前已在许多领域和部门获得了实际的应用,如基础学科领域、工业生产领域、信息技术领域、国防军事领域等。 近年来神经网络控制得到了空前的发展,由于它的许多优异特性,可以在控制系统中采用神经网络对难以精确描述的复杂的非线性对象进行建模,充当控制器或估计器,进行推理及优化计算。对于飞行器姿态及其导航系统来说,用传统的控制方法难以满足精度的要求,因此必须寻找一种有效的控制方法来解决此类的问题,这样神经网络控制显示了它固有的优势。 本文主要研究了神经网络对飞行器姿态的控制及在GPS/INS组合导航系统中的应用问题。首先,讨论了飞行器的姿态预测控制问题,在非线性预测模型中采用BP(Back Propagation)算法,得到了系统的预报输出,针对期望输出的轨迹,利用非线性优化方法极小化二次型性能指标,得到了最优控制序列。然后利用飞行器姿态角方程组实现了对姿态的预测。其次,研究了飞行器姿态控制的问题,在该问题中主要利用BP算法调整飞行器的角速度在坐标系中的投影,在算法中引入了惯性因子,从而使收敛速度加快。为了使系统在实际中具有更强的鲁棒性,对飞行器的姿态角进行了预测输出,仿真显示了飞行器姿态角的变化情况。最后,讨论了基于H_∞滤波的神经网络在GPS/INS组合导航系统中的研究。对于GPS/INS组合导航系统仅利用神经网络的方法无法得到输入输出间的不定映射关系,于是更多还原

【Abstract】 Traditional type of control needs mathematics model about controlled objects and designs controller based on performance indexes. But controlled objects are complex and uncertainty for nonlinear systems, thus traditional methods cannot solve the problems of modeling and control. Neural network imitates human brain systems using, storing and searching information on different degrees. It can approach arbitrary nonlinear function and has a good many merits of combining character, learning and organizing, tolerant fault and association of ideas. So it can solve many problems in nonlinear control fields. At present it obtains practicing application in many fields and branches, such as foundation knowledge, industrial produce, information technology and national defence military affairs etc.Neural network control receives great development in recently years, for its many good merits, it can be used to get modeling and be looked as controller or estimate device. For aircraft attitude and navigation system, the precision cannot be met by using traditional control. So it needs search an efficient method to solve the problem, and the neural network shows inhere superiority.The aircraft attitude control and GPS/INS navigation system are dealt with based on neural network in this thesis. First of all, the problem of predicting aircraft attitude is discussed by using BP algorithm. The predictive output is obtained. To expectation output locus, the optimal controlled sequence is presented by means of the nonlinear optimizer. The problem of attitude prediction is solved based on aircraft attitude equations. Then, the problem of aircraft attitude control is studied. Using BP algorithm,the aircraft angle speed in coordinate projection is adjusted. For it has strong robusness the inertia! gene is introduced to predict the output of attitude. Simulation shows the change of aircraft attitude. At last, the problems of GPS/INS integral navigation system are researched based on H∞ filter neural networks. The mapping relationship between input and output cannot be got if only using neural networks. So the H∞ filter is combined with the neural network to obtain the expectation output of the system, which much more improves the precision of navigation.BP algorithm of neural network is mostly used in the thesis. It can approach arbitrary nonlinear mapping if BP networks have enough latent layers and nodes. It adopts whole means of learning algorithm and has many joint powers, so it has good generalize ability and tolerant fault character. BP network obtains abroad applications in control fields.Simulation results are given at the end of each chapter to show the validity of each result.更多还原