Research on the Lightweight of Natural Language Understanding Model in Electric Power Field

To address the application requirements of indicator question-and-answer (Q&A) in the electricity industry, this paper researches a natural language understanding model for balancing performance, computing resource consumption, and inference time, and proposes a method to incorporate an early st...

Full description

Saved in:
Bibliographic Details
Main Authors: DONG Zengbo, XU Shiyu, CHEN Xi, XU Bo, XIN Rui, ZHANG Pengfei, SONG Hui
Format: Article
Language:zho
Published: Harbin University of Science and Technology Publications 2024-12-01
Series:Journal of Harbin University of Science and Technology
Subjects:
Online Access:https://hlgxb.hrbust.edu.cn/#/digest?ArticleID=2391
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849416737378795520
author DONG Zengbo
XU Shiyu
CHEN Xi
XU Bo
XIN Rui
ZHANG Pengfei
SONG Hui
author_facet DONG Zengbo
XU Shiyu
CHEN Xi
XU Bo
XIN Rui
ZHANG Pengfei
SONG Hui
author_sort DONG Zengbo
collection DOAJ
description To address the application requirements of indicator question-and-answer (Q&A) in the electricity industry, this paper researches a natural language understanding model for balancing performance, computing resource consumption, and inference time, and proposes a method to incorporate an early stopping mechanism in the lightweight process of knowledge distillation. When training the student model with the knowledge of the teacher model, patient early stopping mechanism is added to each layer of encoding, allowing samples of different complexity to be encoded with different layers, reducing overfitting problems that are prone to occur in the student model, and controlling the inference time of the model. On the dataset constructed by the real data in the power industry, experiments show that when the parameters of the student model are reduced to 60% of the teacher model, and the inference time is reduced by nearly 50% , while the accuracy of the model is only reduced by about 2% , maintaining high availability. Compared with the benchmark models, it achieves a better balance in performance and inference time.
format Article
id doaj-art-05ef79a002a841df9e46e720e39512d6
institution Kabale University
issn 1007-2683
language zho
publishDate 2024-12-01
publisher Harbin University of Science and Technology Publications
record_format Article
series Journal of Harbin University of Science and Technology
spelling doaj-art-05ef79a002a841df9e46e720e39512d62025-08-20T03:33:07ZzhoHarbin University of Science and Technology PublicationsJournal of Harbin University of Science and Technology1007-26832024-12-01290614314910.15938/j.jhust.2024.06.014Research on the Lightweight of Natural Language Understanding Model in Electric Power FieldDONG Zengbo0XU Shiyu1CHEN Xi2XU Bo3XIN Rui4ZHANG Pengfei5SONG Hui6Telecommunication Branch State Grid Hebei Electric Power Co. , Ltd. , Shijiazhuang 050081 ,ChinaSchool of Computer Science and Technology, Donghua University, Shanghai 201620 ,ChinaTelecommunication Branch State Grid Hebei Electric Power Co. , Ltd. , Shijiazhuang 050081 ,ChinaSchool of Computer Science and Technology, Donghua University, Shanghai 201620 ,ChinaTelecommunication Branch State Grid Hebei Electric Power Co. , Ltd. , Shijiazhuang 050081 ,ChinaTelecommunication Branch State Grid Hebei Electric Power Co. , Ltd. , Shijiazhuang 050081 ,ChinaSchool of Computer Science and Technology, Donghua University, Shanghai 201620 ,ChinaTo address the application requirements of indicator question-and-answer (Q&A) in the electricity industry, this paper researches a natural language understanding model for balancing performance, computing resource consumption, and inference time, and proposes a method to incorporate an early stopping mechanism in the lightweight process of knowledge distillation. When training the student model with the knowledge of the teacher model, patient early stopping mechanism is added to each layer of encoding, allowing samples of different complexity to be encoded with different layers, reducing overfitting problems that are prone to occur in the student model, and controlling the inference time of the model. On the dataset constructed by the real data in the power industry, experiments show that when the parameters of the student model are reduced to 60% of the teacher model, and the inference time is reduced by nearly 50% , while the accuracy of the model is only reduced by about 2% , maintaining high availability. Compared with the benchmark models, it achieves a better balance in performance and inference time.https://hlgxb.hrbust.edu.cn/#/digest?ArticleID=2391electricity domainnatural language understanding modellightweight modelknowledge distillationearly stopping mechanism
spellingShingle DONG Zengbo
XU Shiyu
CHEN Xi
XU Bo
XIN Rui
ZHANG Pengfei
SONG Hui
Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
Journal of Harbin University of Science and Technology
electricity domain
natural language understanding model
lightweight model
knowledge distillation
early stopping mechanism
title Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
title_full Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
title_fullStr Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
title_full_unstemmed Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
title_short Research on the Lightweight of Natural Language Understanding Model in Electric Power Field
title_sort research on the lightweight of natural language understanding model in electric power field
topic electricity domain
natural language understanding model
lightweight model
knowledge distillation
early stopping mechanism
url https://hlgxb.hrbust.edu.cn/#/digest?ArticleID=2391
work_keys_str_mv AT dongzengbo researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT xushiyu researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT chenxi researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT xubo researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT xinrui researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT zhangpengfei researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield
AT songhui researchonthelightweightofnaturallanguageunderstandingmodelinelectricpowerfield