基于自适应滤波的降阶在线系统辨识

发布时间：2018-12-20 14:38

【摘要】：本文解决的问题是针对存在一定纯延时的高阶复杂系统的在线辨识。如果不去辨识纯延时会导致系统辨识结果的阶数很高,这样不利于控制系统的设计而且会带来非常复杂而又繁琐的计算。因此本文提出了针对延时系统的一种新的降阶办法。本文是以单晶硅的生产过程为背景,以单晶炉控制系统中的变量为数据进行的系统辨识。由于辨识精度的需要,我们将数据进行分为多个辨识周期进行在线辨识。在线辨识算法采用的是在线递推的最小二乘辨识。这种算法计算量较小有利于在系统资源受限的PLC上独立完成。针对实际的辨识过程中所遇到的系统阶数过高的问题,采用间接辨识的手段。先估计输入输出之间存在的延时然后给辨识模型进行在线辨识。采用的降阶办法是基于LMS自适应滤波估计延时的方法,并且在不同信噪比下比较了此方法与几种传统方法的估计效果。在线降阶辨识算法的步骤是：先通过AIC法则辨识出系统的阶数,然后使输入输出数据传递给自适应滤波器确定出系统的纯延时。根据纯延时从而降低系统的阶数。再将系统输入输出以及纯延时这三个变量给在线辨识算法,通过不断在线修正辨识参数来跟踪系统。通过对比估计延时的辨识和没有估计延时的辨识的计算效率以及精确度可以论证此方法的有效性。
[Abstract]:The problem solved in this paper is the online identification of high order complex systems with pure delay. If the pure delay is not identified, the order of the system identification results will be very high, which is not conducive to the design of the control system and will lead to very complex and cumbersome calculation. Therefore, a new order reduction method for delay systems is proposed in this paper. In this paper, the single crystal silicon production process as the background, single crystal furnace control system variables as the data for the identification of the system. Due to the need of identification accuracy, we divide the data into multiple identification cycles for online identification. The online identification algorithm adopts the least square identification of the online recursion. This algorithm is easy to be implemented independently on the PLC with limited system resources. In order to solve the problem that the system order is too high in the actual identification process, indirect identification is used. The delay between input and output is estimated first and then the identification model is identified online. The order reduction method is based on LMS adaptive filter to estimate the delay, and the estimation effect of this method is compared with that of several traditional methods under different SNR. The steps of online order reduction identification algorithm are: first, the order of the system is identified by AIC rule, and then the input and output data are transferred to the adaptive filter to determine the pure delay of the system. The order of the system is reduced according to the pure delay. Then the system input and output variables and pure delay are given to the on-line identification algorithm to track the system by constantly modifying the identification parameters online. The effectiveness of this method can be proved by comparing the computational efficiency and accuracy between the estimated delay identification and the identification without the estimated delay.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN304.12;TN713

【相似文献】