A triple-optimal semi-supervised regression algorithm based on a self-training framework

Abstract

The invention relates to a triple optimization semi-supervised regression algorithm under a self-training framework, which relates to the technical field of semi-supervised regression algorithm. The invention screens the unlabeled samples and the labeled samples, establishes a Gaussian process regression model, predicts the label value of the unlabeled sample set by using the model, and obtains afalse label sample set. Selecting the sample set of pseudo-label satisfying the condition by using the criterion of confidence degree; Further selecting a sample set of false tags with high credibility by judging confidence level, At that same time, the tag sample set and the untag sample set are updated, the untag sample set and the tag sample set are re-screened and the threshold is updated by utilizing the Gaussian process regression model, the false tag sample set of the untag sample set is predicted, the confidence degree of the false tag sample set is judged, and the self-training cycleis entered until the set cycle times are reached. The invention realizes the confidence judgment of the false label sample, and introduces a self-training frame to improve the utilization rate of thelabel-free sample, so as to improve the prediction effect of the model after using the label-free sample.

Description

technical field [0001] The invention relates to the technical field of semi-supervised regression algorithms, in particular to the technical field of three-optimized semi-supervised regression algorithms under the self-training framework. Background technique [0002] Some important quality variables in industrial processes such as chemical industry, metallurgy, and fermentation cannot be measured by online instruments, and there is a serious lag in offline analysis in the laboratory. Therefore, it is necessary to use some sample data that can be directly measured. important quality variables to predict. With the development of science and technology, especially the development of industrial big data technology, unlabeled samples are becoming more and more easy to obtain in large quantities, while the cost of obtaining labeled samples is still high, resulting in few labeled samples in some industrial processes. It is difficult for modeling methods to guarantee the predictio...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that there are few labeled samples and high acquisition costs in the industrial process, and the traditional semi-supervised learning cannot guarantee sufficient and accurate prediction of unlabeled samples, the present invention proposes a three-optimized semi-supervised regression algorithm under the self-training framework

Images