【作者】 沈俊；

【导师】 苏义鑫；

【作者基本信息】 武汉理工大学 ， 控制科学与工程， 2011， 硕士

【摘要】 磁轴承具有无摩擦、寿命长、无需润滑、回转精度高等优点,被广泛应用于航天、机械加工、动力传动、能源交通等领域。磁轴承控制系统是一个涉及了电磁学、控制科学、机械科学以及转子动力学等多门学科的复杂非线性开环不稳定系统。本文研究对象是主动磁轴承,采用基于改进粒子群算法优化的神经网络在线调整PID控制器参数的方式来实现对磁悬浮转子的闭环控制。本文主要做了以下几个方面的工作：阐述磁轴承控制系统的基本构成和工作原理。通过分析磁轴承系统中的电磁关系建立磁轴承控制系统的数学模型。介绍PID控制算法并针对算法的不足提出改进措施。在此基础上提出磁轴承PID控制方案并对设计的磁轴承PID控制器设置相关参数进行仿真研究,在深入分析仿真结果后提出采用BP神经网络在线调整PID控制器参数KP,K1,KD的值。提出磁轴承BP神经网络PID控制器设计方案。根据系统复杂程度设定BP神经网络结构,选取2个输入神经元,15个隐层神经元,3个输出神经元。BP网络的两个输入分别是磁悬浮转子的位移偏差和偏差的变化率,三个输出分别对应PID控制器的三个参数KP,K1,KD。先根据一组输入输出样本数据对BP神经网络进行离线训练,然后BP网络根据系统性能在线调整网络权值系数使得P1D控制器参数最优。针对BP算法的不足引入粒子群算法。通过分析粒子群算法中各个参数对算法性能的影响,提出相应的改进方法,其中重点提出基于动态变异思想的改进粒子群算法。该算法主要对粒子群算法中的惯性权重w进行改进,即先使w线性衰减,再根据粒子收敛的具体情况对惯性权重w进行二次修正。利用改进粒子群算法替代BP算法优化神经网络的权值系数。在此基础上,提出基于改进粒子群算法优化的磁轴承神经网络PID控制方案。仿真结果表明,经改进粒子群算法优化设计的磁轴承神经网络PID控制方案不仅使磁悬浮转子具有优良的动态性能和稳态性能,而且使控制系统具备良好的抗干扰能力。更多还原

【Abstract】 Magnetic bearings with no friction, long life, no lubrication, high turning precision, is widely used in aerospace, mechanical processing, power transmission, energy, transportation and other fields. Magnetic bearing control system is an equipment involved electromagnetic, control theory, mechanical theory, rotor dynamics and other subjects of the complex non-linear open-loop unstable system. The research object is the active magnetic bearings. In this paper, improved particle swarm optimization is used to optimize neural network and adjust the PID controller parameters online to achieve closed-loop control of magnetic suspension rotor. In this paper, I do the following work.Magnetic bearing control system is described the basic structure and working principle. The mathematical model of Magnetic bearing system is established by analyzing the relationship between the electromagnetic magnetic bearing control system. The disadvantage of PID control algorithm is described and the improvement measures are proposed. The magnetic bearing PID controller design is proposed on the base of PID algorithms. Then the program is set right parameters to simulation. Deeply analysis the simulation results presented by BP neural network PID controller.It is proposed a magnetic bearing BP neural network PID controller design. On the basis of the complexity of the system to set BP neural network structure, select 2 input neurons,15 hidden layer neurons, three output neurons. The two inputs are the displacement of the rotor magnetic deviation and deviation the rate of change, three outputs are corresponding to the three parameters of PID controller. A group of sample based on input s and outputs data is used to train BP neural network. BP network based on system performance and then adjust the network weights coefficient line to set the optimal PID controller parameters.The particle swarm optimization is introduced to solve the advantage of the BP algorithm. Particle swarm algorithm is improved by analyzing the various parameters on algorithm performance, the corresponding improved method, which highlight the dynamic variation of ideas based on improved particle swarm optimization. The algorithm is mainly about the particle swarm algorithm to improve the weight of inertia, that is, firstly to adjust as the linear attenuation, according to the specific circumstances of the convergence of particle inertia weight for the second amendment. Alternative use of improved particle swarm optimization neural network BP algorithm weight value. On this basis, improved particle swarm optimization based on the magnetic bearing neural network PID control scheme is proposed.Simulation results show that the improved particle swarm optimization design of magnetic bearing PID neural network control scheme not only has excellent magnetic rotor dynamic performance and steady-state performance, but also the control system with good anti-jamming capability.更多还原