ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs

Large language models (LLMs) have recently demonstrated notable performance, particularly in addressing the challenge of extensive data requirements when training traditional forecasting models. However, these methods encounter significant challenges when applied to high-dimensional and domain-speci...

Full description

Saved in:
Bibliographic Details
Main Authors: Jinglei Pei, Yang Zhang, Ting Liu, Jingbin Yang, Qinghua Wu, Kang Qin
Format: Article
Language:English
Published: MDPI AG 2025-04-01
Series:Machine Learning and Knowledge Extraction
Subjects:
Online Access:https://www.mdpi.com/2504-4990/7/2/35
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849431854952742912
author Jinglei Pei
Yang Zhang
Ting Liu
Jingbin Yang
Qinghua Wu
Kang Qin
author_facet Jinglei Pei
Yang Zhang
Ting Liu
Jingbin Yang
Qinghua Wu
Kang Qin
author_sort Jinglei Pei
collection DOAJ
description Large language models (LLMs) have recently demonstrated notable performance, particularly in addressing the challenge of extensive data requirements when training traditional forecasting models. However, these methods encounter significant challenges when applied to high-dimensional and domain-specific datasets. These challenges primarily arise from inability to effectively model inter-variable dependencies and capture variable-specific characteristics, leading to suboptimal performance in complex forecasting scenarios. To address these limitations, we propose <b>ADTime</b>, an adaptive LLM-based approach for multivariate time series forecasting. ADTime employs advanced preprocessing techniques to identify latent relationships among key variables and temporal features. Additionally, it integrates temporal alignment mechanisms and prompt-based strategies to enhance the semantic understanding of forecasting tasks by LLMs. Experimental results show that ADTime outperforms state-of-the-art methods, reducing MSE by 9.5% and MAE by 6.1% on public datasets, and by 17.1% and 13.5% on domain-specific datasets. Furthermore, zero-shot experiments on real-world refinery datasets demonstrate that ADTime exhibits stronger generalization capabilities across various transfer scenarios. These findings highlight the potential of ADTime in advancing complex, domain-specific time series forecasting tasks.
format Article
id doaj-art-bfe7620103b44be28a94e4ff5c21a8ec
institution Kabale University
issn 2504-4990
language English
publishDate 2025-04-01
publisher MDPI AG
record_format Article
series Machine Learning and Knowledge Extraction
spelling doaj-art-bfe7620103b44be28a94e4ff5c21a8ec2025-08-20T03:27:29ZengMDPI AGMachine Learning and Knowledge Extraction2504-49902025-04-01723510.3390/make7020035ADTime: Adaptive Multivariate Time Series Forecasting Using LLMsJinglei Pei0Yang Zhang1Ting Liu2Jingbin Yang3Qinghua Wu4Kang Qin5Institute of Computing Technology, University of Chinese Academy of Sciences, Beijing 100190, ChinaResearch Institute of Petroleum Processing, SINOPEC, Beijing 100083, ChinaInstitute of Computing Technology, University of Chinese Academy of Sciences, Beijing 100190, ChinaInstitute of Computing Technology, University of Chinese Academy of Sciences, Beijing 100190, ChinaInstitute of Computing Technology, University of Chinese Academy of Sciences, Beijing 100190, ChinaResearch Institute of Petroleum Processing, SINOPEC, Beijing 100083, ChinaLarge language models (LLMs) have recently demonstrated notable performance, particularly in addressing the challenge of extensive data requirements when training traditional forecasting models. However, these methods encounter significant challenges when applied to high-dimensional and domain-specific datasets. These challenges primarily arise from inability to effectively model inter-variable dependencies and capture variable-specific characteristics, leading to suboptimal performance in complex forecasting scenarios. To address these limitations, we propose <b>ADTime</b>, an adaptive LLM-based approach for multivariate time series forecasting. ADTime employs advanced preprocessing techniques to identify latent relationships among key variables and temporal features. Additionally, it integrates temporal alignment mechanisms and prompt-based strategies to enhance the semantic understanding of forecasting tasks by LLMs. Experimental results show that ADTime outperforms state-of-the-art methods, reducing MSE by 9.5% and MAE by 6.1% on public datasets, and by 17.1% and 13.5% on domain-specific datasets. Furthermore, zero-shot experiments on real-world refinery datasets demonstrate that ADTime exhibits stronger generalization capabilities across various transfer scenarios. These findings highlight the potential of ADTime in advancing complex, domain-specific time series forecasting tasks.https://www.mdpi.com/2504-4990/7/2/35time seriestime series forecastinglarge language models
spellingShingle Jinglei Pei
Yang Zhang
Ting Liu
Jingbin Yang
Qinghua Wu
Kang Qin
ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
Machine Learning and Knowledge Extraction
time series
time series forecasting
large language models
title ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
title_full ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
title_fullStr ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
title_full_unstemmed ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
title_short ADTime: Adaptive Multivariate Time Series Forecasting Using LLMs
title_sort adtime adaptive multivariate time series forecasting using llms
topic time series
time series forecasting
large language models
url https://www.mdpi.com/2504-4990/7/2/35
work_keys_str_mv AT jingleipei adtimeadaptivemultivariatetimeseriesforecastingusingllms
AT yangzhang adtimeadaptivemultivariatetimeseriesforecastingusingllms
AT tingliu adtimeadaptivemultivariatetimeseriesforecastingusingllms
AT jingbinyang adtimeadaptivemultivariatetimeseriesforecastingusingllms
AT qinghuawu adtimeadaptivemultivariatetimeseriesforecastingusingllms
AT kangqin adtimeadaptivemultivariatetimeseriesforecastingusingllms