On-line soft measurement modeling method for 4-CBA content based on least square algorithm of restricted memory part

Abstract

This invention discloses a modeling method for on-line measuring 4-CBA contents based on limited -memory section of method of least square, which chooses nine most important process variables to 4-CBA content as assistance variables of flexible measuring model. And it employs the rolling correcting samples method of averaging and variance to bring information of old samples into the model to avoid missing information. The advantages of the invention are the following: 1, 4-CBA flexible measuring model is of linear model and of simple structure; 2, 4-CBA flexible measuring model is set up through rolling pattern with self-adaptation for systematic slow time-varying characteristics; 3, It employs l limited -memory section of method of least square to overcome saturation of sample; 4, the flexible measuring model of 4-CBA content is dynamic model with high accuracy.

Description

technical field [0001] The invention relates to an online soft sensor modeling method for 4-CBA content based on limited memory partial least squares algorithm. Background technique [0002] PTA (Prified Terephthalic Acid) is one of the main raw materials for the production of polyester fibers. figure 1 It shows the basic technological process of p-xylene oxidation reaction by AMOCO method. p-xylene, acetic acid, mother liquor and catalyst cobalt, manganese and bromine are stirred evenly in the mixing tank, and the mixed solution is sent into the oxidation reactor and simultaneously into the oxidation reactor. The reactor also contains air. The oxidation process is carried out in two stages, the first stage takes place in the oxidation reactor and the second stage takes place in the first crystallizer. The reacted liquid is depressurized in the subsequent three crystallizers, condensed and then filtered and dried to obtain terephthalic acid (TA) and p-carboxybenzaldehyde (...

AI Technical Summary

Problems solved by technology

[0008] 1) The mechanism modeling method can only be applied to industrial processes with clear process mechanism. Once the process is complex and the process mechanism is not completely clear, this method is no longer applicable

[0009] 2) The empirical regression model is essentially a static model, without considering the dynamic characteristics of the object

At the same time, compared with the linear model, the structure is complex

[0010] 3) The neural network modeling method needs to collect a large number of training samples in advance, the structure is complex, the training time of the model is long, it is not conducive to online modeling, and the promotion ability of the model is poor

[0011] 4) The partial least squares modeling method and the principal component analysis method are only suitable for the modeling of steady systems. If the process object has slow time-varying characteristics, the original model will no longer be applicable

[0012] 5) Although the limited memory least squares modeling method can realize the modeling of objects with slow time-varying characteristics to a certain extent, the algorithm requires a small degree of correlation between input variables. If the correlation between input variables Larger (that is, with multicollinearity), the error of the model obtained by the limited memory least squares algorithm is larger

Images