【作者】 王婷；

【导师】 靳其兵；

【作者基本信息】 北京化工大学 ， 控制科学与工程， 2011， 硕士

【摘要】 精确的对象数学模型是先进控制理论应用的必要前提,是否能够得到表征系统特性的模型对优化控制起至关重要的作用。本文对系统辨识的原理方法、信号的选择等方面重点介绍,并通过仿真采集数据信息进行辨识计算。针对实际工业应用情况做了以下几个方面的工作：1、介绍系统辨识的发展情况以及现代辨识方法,描述了模型类型、建模方法以及误差准则的选取。对经典的辨识算法：最小二乘法、图解法、基于FIR模型的最小二乘法通过仿真对系统模型进行辨识得出各算法的优缺点。2、研究NLJ、粒子群优化算法(PSO)以及遗传算法(GA)在系统辨识中的应用,并针对在实际应用中,遗传算法收敛速度慢、精度较低、易陷入局部最优等缺点,通过修改参数变化范围的上限对遗传算法进行改进,能够保证算法跳出局部最优所搜到参数的无偏一致估计。在满意度的概念上利用改进的遗传算法优化控制器参数,能够得到满足要求的控制系统。3、针对实际应用过程中,控制回路不允许转换成开环形式但经典辨识算法无法直接应用于闭环辨识的情况,将粒子群算法的全局搜索能力和Rosenbrock算法的局部搜索能力结合,提出了PSO-Rosenbrock算法。该算法不需要控制器的先验知识,在闭环条件下,对任意测试信号都能获得待估对象的所有参数,不仅提高了收敛速度,缩短了辨识时间,同时极大地减小了模型辨识参数对参数初始值依赖性。4、介绍常用的多项式预测滤波、中值滤波以及三次函数替代滤波算法,均具有较强的滤波能力。但是过于平滑的或有失真的波形会减少数据提供的有效信息,降低辨识精度。通过对中值滤波数据采用取均值的方式进行滤波,能够有效地去除脉冲废值并提供更多的信息量,与单纯中值滤波相比辨识结果大幅改善。更多还原

【Abstract】 Application of advanced control theory is based on object’s accurate mathematical model and the system character plays a critical role in optimal control. In this paper, the principle of system identification methods, signal selection and so on are focused and make identification calculations based on collected data information get by simulation. Based on practical industrial application, this paper make following contributions:1、Introduce the development of system identification as well as modern methods and describe the model types, modeling methods and error criteria. Analysis classical identification algorithms:least squares method, graphical method, direct identification method for continuous by simulation and obtain the advantages and disadvantages of each method.2、Research NLJ, particle swarm optimization (PSO) and genetic algorithm (GA) in system identification applications. In real applications, GA which is easy to fall into local optimum has low convergence speed, less precise. In this paper, GA is improved by adding high limit which ensure that the algorithm can jump out of local optimal search parameters to get consistent and unbiased estimates. Based on the concept of satisfaction, optimize the controller parameters by using the improved genetic algorithm meet the requirements of the control system.3、For the practical application process, control loop is not allowed converted into the open-loop form, but classical identification algorithms can not be applied to the closed-loop identification directly. a novel identification method—PSO-Rosenbrock is proposed by integrating global identification ability of Particle Swarm Optimization (PSO) and local search competence of Rosenbrock. The algorithm does not require prior knowledge of controllers on closed loop conditions and can obtain the all of the parameters to be estimated based on arbitrary test signal. This algorithm can not only improve the convergence rate but also reduce the dependence of identification parameters on initial parameters.4、Describe commonly used polynomial prediction filter, median filter and three alternative filtering algorithms function which all have strong filtering. But if the waveform is too smooth or anamorphic it will reduce useful information provided by data and identification accuracy. Take average valued of median filter can effectively remove the scrap value of the pulse and provide more information. Compared with the simple median filter, this method can improve identification results.更多还原