TimeGPT-based multi-step-ahead key quality indicator forecasting for industrial processes

Xinmin Zhang, Yuwei Chen, Bocun He, Zhihuan Song, Manabu Kano

Abstract

Multi-step prediction is one of the most challenging problems in the field of industrial soft sensing. Recently, large language models have been widely used in various fields. Inspired by TimeGPT, a popular large-scale model for time series forecasting, this paper proposes a novel multi-step key quality indicator forecasting method for industrial processes, namely the TimeGPT-based Multi-step-ahead Forecasting (TiMF) model. The proposed TiMF model is designed based on pre-trained TimeGPT, and historical process variable information is integrated into the prediction model as an auxiliary guide to improve the utilization of industrial data information. To evaluate the effectiveness of the proposed method, it was applied to the debutanizer industrial process and the sintering industrial process. The application results show that the proposed TiMF can achieve better prediction accuracy than other existing methods. This work provides a new attempt for the industrial soft sensing application of the large-scale time series prediction model. • A TimeGPT-based multi-step-ahead quality prediction model, TiMF, was developed. • TiMF introduces auxiliary variables as guidance to improve quality prediction tasks. • TiMF is more suitable for soft sensing applications in multivariate industrial processes. • TiMF achieved better prediction performance than conventional methods.
BibTeX
@inproceedings{zhang2025timegpt,
  author = {Xinmin Zhang and Yuwei Chen and Bocun He and Zhihuan Song and Manabu Kano},
  title = {{TimeGPT-based multi-step-ahead key quality indicator forecasting for industrial processes}},
  booktitle = {Control},
  year = {2025},
  doi = {10.1016/j.conengprac.2025.106410},
}

Loading related content...

Figure