A Meta-Invariant Feature Space Learning Method for Cross-Domain Prediction

Abstract

A meta-invariant feature space learning method for cross-domain prediction, which is characterized by using existing data as source domain data, and grouping and pairing the source domain data; establishing a prediction model for each pair of data, and then constructing the pairing The invariant feature space learning model of the data learns the invariant feature space of each paired data through collaborative training; the invariant feature space learning model is used as the base model, and the meta-invariant feature space between different pairs is learned through the meta-learning method, The meta-invariant feature space learning model is obtained, and then the target domain is predicted based on the meta-invariant feature space learning model. The invention obtains the invariant feature space under the two working conditions through collaborative learning, and solves the problem of marginal distribution adaptation. Improved prediction accuracy across domains.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, in particular to a cross-domain prediction method oriented to intelligent manufacturing, in particular to a meta-invariant feature space learning method for cross-domain prediction. Background technique [0002] Cross-domain prediction is an important research issue in the field of machine learning. In the field of manufacturing, the main reason is that due to the large changes in working conditions, the marginal distribution and conditional distribution of data are quite different, and it is difficult to adapt the model trained on the source domain to new conditions. Working conditions, a typical problem in the manufacturing field is tool wear prediction. Real-time monitoring of tool wear is of great significance to the dynamic control of the machining process, especially for complex aircraft parts that use a large number of difficult-to-machine materials. The tool wear during the ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to address the problem of cross-domain prediction, and propose a meta-invariant feature space learning method, which takes the invariant feature space learning model as the base model, learns the meta-invariant feature space through the meta-learning method, and then based on the meta-invariant Variational Feature Space Learning Models for Prediction of Target Domains

Images