Industrial high-order dynamic process soft measurement method based on multi-hidden-layer weighted dynamic model

Abstract

The invention discloses an industrial high-order dynamic process soft measurement method based on a multi-hidden-layer weighted dynamic model. According to the method, a sliding window is introduced,and a multi-hidden-layer dynamic model is established in each sliding window for each group of online samples, so the local autocorrelation of data in a hidden space and the high-order dynamic relation of hidden variables in time sequence are fully considered, and the description of the data can be more accurate; in combination with a support vector data description method, a global weight of an online sample is calculated, and the multi-hidden-layer weighted dynamic model is established; after the parameters of the model are obtained, a locally weighted linear regression model is establishedso as to obtain a quality variable estimation value of the online sample.

Description

technical field [0001] The invention belongs to the field of industrial non-stationary process soft sensor modeling and application, in particular to an industrial high-order dynamic process soft sensor modeling and online estimation method based on a multi-hidden layer weighted dynamic model. Background technique [0002] The industrial production process is very complicated, and it is difficult to achieve real-time measurement to obtain real-time data in the entire production process. Soft measurement refers to estimating the value of the quality variable by establishing a mathematical model between the quality variable and the process variable that is easy to measure. [0003] At present, the common soft sensor models include: partial least squares regression, decision tree, neural network, etc., but the model is relatively simple, there is noise in the data, the data distribution is non-Gaussian, the lack of quality variables, and the process is dynamic and other characte...

AI Technical Summary

Problems solved by technology

[0003] At present, the common soft sensor models include: partial least squares regression, decision tree, neural network, etc., but the model is relatively simple, there is noise in the data, the data distribution is non-Gaussian, the lack of quality variables, and the process is dynamic and other characteristics are difficult to solve. More complex mathematical models need to be further established

[0004] In the above problems, it is usually difficult to deal with the dynamic characteristics of the process. Common dynamic models such as linear dynamic systems can establish the dynamic characteristics of the observed data hidden in time, but the description of the dynamic characteristics hidden in time is relatively simple, and there is no It is a good description of the high-order dynamic characteristics of the data in time. When the hidden dynamic characteristics of the data are related to the data of several previous moments, the model is often difficult to describe well

Images