Industrial soft measurement method based on multiple NPE (neighborhood preserving embedding) regression models

Abstract

The invention discloses an industrial soft measurement method based on multiple NPE (neighborhood preserving embedding) regression models. The method aims to comprehensively mine the local neighboring characteristic relation of data and establish multiple regression models to realize on-line soft measurement of product quality. Specifically, the method comprises steps as follows: firstly, distance neighborhood, time neighborhood and angle neighborhood are searched for each sample point of input data; then, a distance NPE regression model, a time NPE regression model and an angle NPE regression model are established correspondingly; next, an estimation value for output by each NPE regression model is taken as input again, and regression models between the estimation values and the output are established again by use of a partial least-squares algorithm; finally, the established regression models are used for on-line soft measurement. The method is a preferred soft measurement implementation scheme due to the fact that more potential useful information can be more comprehensively mined and the generalization ability of the soft measurement models is guaranteed by use of the multiple regression models during soft measurement.

Description

Technical field [0001] The invention relates to a soft measurement method, in particular to an industrial soft measurement method based on multi-nearest neighbor preserving embedded regression model. Background technique [0002] Stable and continuous production of qualified products is the basic way to ensure corporate profits. Real-time monitoring of product quality indicators is therefore essential in the entire process control system. In the production process, it is usually desired to measure effective information about product quality in real time. If some quality index information cannot be directly measured, other indexes that can directly reflect product quality information will be measured indirectly. Generally speaking, instruments and equipment that measure quality indicators online in real time are more expensive and costly to maintain later than instruments that measure pressure, temperature, and flow. If offline analysis equipment is used, it is impossible to mea...

AI Technical Summary

Problems solved by technology

Therefore, although the NPE method has started to be used in soft sensing, it still needs further research and development in mining the local neighbor relationship of the data.

To put it simply, NPE simply considers the local structural features with similar spatial distances, and does not consider the local features of the sampling time neighbors and the spatial angle neighbors when establishing the soft sensor model.

In addition, the soft sensor accuracy that can be achieved by a single NPE soft sensor model is limited, and it does not have the strong generalization ability of multiple soft sensor models.

Therefore, the use of NPE to establish a regression model to implement soft sensing remains to be further studied

Images