【作者】 冯琳琳；

【导师】 高巍；

【作者基本信息】 吉林大学 ， 系统工程， 2008， 硕士

【摘要】 随着航天技术的发展,人造卫星的挠性问题越来越突出,人们对卫星的姿态控制要求越来越高。为探索用于挠性卫星姿态控制的新方法,本文针对挠性卫星姿态稳定控制进行了基于改进遗传算法的优化模糊神经网络辨识器和控制器的设计研究。建立了带有太阳帆板的三轴稳定卫星姿态运动动力学模型。分析实值编码较二进制编码的优势和特点,提出对标准实值编码遗传算法的改进方法,使之优化性能大大提高。将改进GA应用于模糊神经网络的结构参数优化,加快了模糊神经网络的拟合速度与精度。研究了挠性卫星姿态基于遗传算法的模糊神经网络控制器的设计方法,用改进GA对Takagi-Sugeno型模糊神经网络辨识器均值、方差和权值及控制器的权值同时进行优化设计,离线训练时应用模糊C-均值聚类对输入角度进行聚类,确定控制器的均值和方差。优化后模糊神经网络辨识器和控制器的性能较纯模糊神经网络辨识器和控制器更加突出。MATLAB仿真结果证明所设计的辨识器和控制器均能满足挠性卫星姿态控制性能要求。更多还原

【Abstract】 The satellites more and more play a major role in national economy and military affairs. A lot of business can not get away man-made satellites, for instance, correspondence, navigation, weather, and so on. With the development of the aerospace technology, the man-made satellites construction is increasingly complexity, so the flexibility problem of satellites impacting on attitude of satellite increasingly stands out, and the requirements for the flexible satellite attitude control are getting more and more complex. The flexibility structure of the satellite result in a few accidents of the control performance descend or astaticism, which indicates both classical control and modern control havn’t completely solved these issues, but intelligent control basically independent of model of controlled object. The study on this line not only has certain help for design of vehicle control systems of current type satellites, but also momentous sense on control system design of large-scale space structures such as manned vehicles and space stations.In order to study new methods used in satellite attitude control, this thesis presents the identification and controller of T-S type fuzzy neural network based on genetic algorithm. Then this thesis makes simulation study on the attitude stabilization control of flexible satellite and analyses control outcome.Firstly, the models of the three-axis stabilized attitude control systems of a rigid satellite and a flexible satellite with a pair of solar arrays are set up. The model is composed of kinematics equation and dynamics equation. The model of the attitude control of the three-axis satellite is decoupled under assumer and predigestion, that is to say, the model is used in simulation.Secondly, the simple genetic arithmetic (SGA) is improved. Genetic algorithms are search algorithms based on the mechanics of natural selection and natural genetics. In general, a solution of an objective function is represented as a chromosome in a string structure, and each element represents a parameter in the solution. Through a search process such as selectness, crossover and mutation operators, the choice individuals in the string structure increase with series, and finally the chromosomes will approach the global optimum. The solutions are commonly encoded as binary strings, algorism strings and Gray Code, but in dimensions and high precision continuum problem, the second is better than the first, so in this thesis, genetic algorithms uses algorism coding. Genetic operations of simple genetic algorithms (SGA) easily decrease diversity of genetic population, reduce competition of individuals and result in premature, so genetic algorithms can’t find global optimum. Aimed at these problems, the bettered genetic algorithms (BGA) is put forward. It changes the roulette wheel selection into the selection based on the fitness of individuals in population in order to preserve the individuals with the high fitness. The selected individuals will be put into the second process. Used changing P_c and P_m to adapt the changing situation. It can increase diversity of the population and avoid premature. In the end, BGA is tested by De Jong function. The result shows the feasibility and effectiveness of BGA in overcoming premature and improving convergence speed and accuracy.Thirdly, There were three input/ three output Neuro-Fuzzy network for identification. And there were three Fuzzy congregation of gauess function. Used BGA to optimize the typical value, variance and weighting of Neuro-Fuzzy network. The Neuro-fuzzy network combined the structural knowlage of Fuzzy reasoning and learning capability of neural nerwork. It combined the two field perfectly. It can solve the problems without the precise model, only use the user and the perfeson’s experience. But because of the experience and the parameter of network were limited, it is hard to make the network model. So we used the beterred genetic algorithms to optimize the structure of neural network. We verify the feasibility and validity by using BGA to optimize the Neuro-fuzzy network to simulate the identification of a three-axis-stabilized Flexible satellite.Therefore, this paper presents the T-S type Neuro-Fuzzy control system of a three-axis-stabilized Flexible satellite. From the identification to get the change rate of angle, It will help us to amend the learning rate and parameter of the controller. It was made five subfunction for the three axes angle and it’s rate of the Flexible satellite. And the subfunction is gauss function. The control moment of the flywheel is between [-0.2,0.2]. I got the center of the clustering with fuzzy C-mean clustering, and got the typical value of the Neuro-fuzzy controller. By the back structure, we can got the weighting of the Neuro-fuzzy contoller. After the clustering, a few parameter will be optimized, It will be faster. And the adaptation function is J(ITAE). It will accurate to use the parameter K to adjust the learning rate. This paper simulate the three axis angle of the flexible satellite, when the inertia matrixes rise from 70% to 100%. The result showd that the perturbation and the vibration of subassembly can be keep down, and the control is precise, after the Neuro-fuzzy to be optimized.At last, we sum the achievement up and pointed out the deficiency in this paper.更多还原