Multi-class imbalance fault classification method based on reinforcement learning and knowledge distillation

Abstract

The invention discloses a multi-class imbalance fault classification method based on reinforcement learning and knowledge distillation, which combines algorithms such as hierarchical clustering, knowledge distillation and reinforcement learning and is used for solving the problem of multi-class unbalanced fault classification. For a multi-class fault classification problem, firstly, for the characteristics that homogeneous class samples have similarities and heterogeneous class samples have large differences in an imbalance problem, hierarchical clustering is used to cluster multiple classes into several clusters, and student networks are established according to different clusters to perform fine-grained classification; and then global information is considered by using a knowledge distillation method, and finally, sample weights are iteratively learned in combination with reinforcement learning, so that an unbalanced fault classification effect is improved. In the process, a reasonable reward function needs to be designed to be matched with a fine-grained knowledge distillation classifier to optimize the sample weight. Compared with other comparison methods, the method provided by the invention has good effect and applicability.

Description

technical field [0001] The invention belongs to the field of industrial process monitoring, in particular to a multi-category unbalanced fault classification method based on reinforcement learning and knowledge distillation. Background technique [0002] In machine learning or deep learning classification, the imbalance of the number of class samples is a very common problem, which exists in various fields, such as bioinformatics, smart grid, medical imaging, fault diagnosis. Most existing classification methods are based on the assumption that the underlying distribution of observed data is relatively balanced. However, real industrial datasets usually violate this assumption and present a skewed distribution or even an extremely unbalanced distribution of class sample sizes. For example, data-driven fault classification is an important part of industrial process monitoring, which exhibit unbalanced skewed distributions due to different frequencies of fault occurrence. In...

AI Technical Summary

Problems solved by technology

However, real industrial datasets often violate this assumption and exhibit skewed distributions or even extremely unbalanced distributions of class sample sizes

Images