Semi-supervised learning industrial process soft measurement modeling method based on evolutionary optimization

Abstract

The invention discloses a semi-supervised learning industrial process soft measurement modeling method based on evolutionary optimization. The method is used for solving the problem of low predictionperformance precision of a traditional soft measurement model caused by less label data acquisition and rich label-free data in the industrial process. Pseudo-label estimation is carried out on randomly selected samples in unlabeled data in an evolutionary optimization mode, the obtained pseudo-label data is added into the labeled data, hybrid modeling is carried out through Gaussian process regression, and the performance of a traditional soft measurement model is effectively improved. Besides, small-batch random selection optimization is performed on the label-free data so as to bring diversified pseudo label data optimization results, and the prediction accuracy and stability of the model can be further enhanced through the idea of fusion ensemble learning so that control and monitoringof the industrial process are enabled to be safer and more reliable.

Description

technical field [0001] The invention relates to the field of industrial process control, in particular to an evolutionary optimization-based semi-supervised learning industrial process soft sensor modeling method. Background technique [0002] With the rapid breakthrough of modern information technology, process industries such as chemical industry, petroleum, medicine and metallurgy have developed rapidly in the direction of scale expansion, process complexity and energy conservation and environmental protection, limited by technical or economic factors, such as instrument testing Conventional detection methods such as off-line analysis and off-line analysis have been unable to achieve on-line measurement of key process variables to meet production needs. As an important indirect measurement method, soft sensor technology has become the basis and key to the successful implementation of various advanced automation technologies in recent years. [0003] The process data requ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem that the model performance of traditional soft sensor modeling in industrial process application is limited by the scarce labeled data, the present invention proposes a semi-supervised learning industrial process soft sensor modeling method based on evolutionary optimization, which can make full use of The beneficial information of labeled data and unlabeled data can effectively improve the above problems and realize the online estimation of key parameters in industrial processes

Images