A Soft Sensor Method Based on Optimal Selection and Optimal Regression of Orthogonal Components

Abstract

The invention discloses a soft sensor method based on the optimal selection and optimal regression of orthogonal components, aiming at optimally selecting the orthogonal components that are beneficial to the prediction quality index, and using the selected orthogonal components to establish a soft sensor for optimal regression. Measurement model. Specifically, the method of the present invention at first utilizes principal component analysis (PCA), independent component analysis (ICA) and partial least squares (PLSR) algorithms to obtain corresponding orthogonal components respectively at first, and then utilizes the nearest neighbor component based on genetic algorithm The optimal choice of orthogonal components is analyzed, and the optimal regression modeling based on the particle swarm optimization algorithm is implemented most effectively using the selected orthogonal components. Compared with the traditional method, the method of the present invention selects the feature components that are beneficial to the prediction quality index in an optimized manner and obtains the final quality index prediction value through the optimal regression vector. Therefore, the soft sensing performance of the method of the present invention is fully guaranteed, and it is a more preferred soft sensing method.

Description

technical field [0001] The invention relates to an industrial soft-sensing method, in particular to a soft-sensing method based on optimal selection and optimal regression of orthogonal components. Background technique [0002] Maintaining the stability of an enterprise's product quality is the fundamental way to improve its market competitiveness and brand effect. These key variables that can directly or indirectly reflect product quality are usually obtained using online analysis methods or offline assay analysis methods. However, on-line analytical instruments are expensive and costly to maintain; while off-line assays require a long time to measure the corresponding data, resulting in a serious lag and unable to reflect the current quality status in a timely manner. In order to obtain product quality information in real time at low cost, soft sensor technology emerges as the times require. The basic idea is to use other easily measurable process variables related to the...

AI Technical Summary

Problems solved by technology

There are few research documents or patents that comprehensively consider the simultaneous use of these three types of algorithms, and they only stay at the level of improving a single algorithm

Secondly, not all of the orthogonal components extracted by PCR, ICR, and PLSR are beneficial to the prediction of quality index data, and there is no feasible solution for how to optimally select the orthogonal components that are helpful for quality prediction

Finally, when using orthogonal components to directly predict quality indicators, the common idea is the least squares regression method, which fails to find the relationship between input and output from the perspective of optimizing the regression coefficient vector

Images