A Reduced Order Online System Identification Based On Adaptive Filtering

This paper is to solve the problem of online system identification.The actual plant,hower,is very complicate and has long pure delay and high order. If we don’t estimate pure time delay,as the result,the order of the system identification is very high.This high order system will bring large amount of caculation and is bad for designing control system.Therefore,a new method of order reduction has been put forward in this paper.The background of this paper is the productive process of monocrystalline silicon while the data of the experiment is the variable in the control system.In order to ensure the accuracy of the identification,we divide the data into several part and identify the plant on the line.The online identification algorithm in this paper is least squares identification.Because of the smaller amount of computations,this algorithm can be easily transplant on the platform of PLC. Sometimes we should resort indirect measures when the system order is too high.The indirect maesure is that estimateting the delay between the input and the output and then identifying the system.To estimate the delay we adopt the mathod of LMS adaptive filter and we also compare the different traditional methods in different signal-to-noise ratio.The step of online identification is that:Firstly,identify the order of the plant by mean of AIC.Secondly,put the data into the LMS adaptive filter to estimate the pure delay,then according to the delay to reduce the order.At last put the data and the delay into identification system and corrected parameter constantly to track the plant.By mean of comparing the mathod of estimating delay identification and the mathod of direct identification, we can prove the effective of this mathod.