A method for predicting diesel yield of hydrocracking unit based on time-lag-aware transformer

RUAN Zhijian; LI Jiahe; LI Jiayang

doi:10.11949/0438-1157.20260582

您当前的位置：

首页 >

文章列表页 >

A method for predicting diesel yield of hydrocracking unit based on time-lag-aware transformer

更新时间：2026-06-26

- A method for predicting diesel yield of hydrocracking unit based on time-lag-aware transformer
- CIESC Journal (2026)
- 作者机构：
  
  蓝卓数字科技有限公司，浙江杭州 310053
- 作者简介：
- 基金信息：
- DOI：10.11949/0438-1157.20260582
  CLC： TP183;TE624
- Received：24 April 2026，
  
  Revised：2026-06-26，
  
  Accepted：26 June 2026，
- 稿件说明：
移动端阅览
RUAN Zhijian, LI Jiahe, LI Jiayang. A method for predicting diesel yield of hydrocracking unit based on time-lag-aware transformer[J/OL]. CIESC Journal, 2026.
DOI：

RUAN Zhijian, LI Jiahe, LI Jiayang. A method for predicting diesel yield of hydrocracking unit based on time-lag-aware transformer[J/OL]. CIESC Journal, 2026. DOI： 10.11949/0438-1157.20260582.

摘要

柴油收率受原料、催化剂及操作参数的非线性、异构性与时滞效应耦合影响，传统数据驱动模型因忽略物理可解释的时序依赖而精度受限。针对上述问题，提出一种基于时滞感知Transformer（Time-Lag Aware Transformer，TLAT）的柴油收率预测方法：首先基于条件互信息量化各变量对收率的有效时滞窗口；继而在Temporal Fusion Transformer基础上，设计工艺先验引导的结构化注意力掩码，约束模型仅在合理时滞范围内建模动态依赖；同时引入静态-动态门控融合模块，利用静态上下文生成可学习门控信号，动态调节动态特征的权重，实现静态变量对动态特征的自适应调制。基于某石化企业140万吨/年加氢裂化装置近3年104维高频率运行数据，TLAT在测试集上MAE达0.0182、R²为0.937，显著优于XGBoost、LSTM、Informer和标准TFT，为智能工厂精益管控提供有效支撑。

Abstract

Diesel yield is jointly influenced by the nonlinear

heterogeneous

and time-lag effects of feedstock properties

catalyst performance

and operational parameters. Conventional data-driven models suffer from limited prediction accuracy due to their inability to explicitly capture physically interpretable temporal dependencies. To address this issue

a diesel yield prediction method based on a Time-Lag Aware Transformer (TLAT) is proposed. First

conditional mutual information is employed to quantitatively identify the effective time-lag window of each process variable with respect to diesel yield. On this basis

a process-prior-guided structured attention mask is constructed within the Temporal Fusion Transformer (TFT) framework

which constrains the model to learn dynamic dependencies only within physically reasonable time-lag ranges. In addition

a static–dynamic gating fusion module is introduced

where static contextual information is utilized to generate learnable gating signals that dynamically adjust the contribution of dynamic features

enabling adaptive modulation of dynamic representations by static variables. The proposed TLAT model is validated using 104-dimensional high-frequency operational data collected over three years from a 1.4 Mt/a hydrocracking unit of a petrochemical enterprise. Experimental results demonstrate that TLAT achieves a mean absolute error (MAE) of 0.0182 and a coefficient of determination (R²) of 0.937 on the test set

significantly outperforming benchmark models including XGBoost

LSTM

Informer and the standard TFT. These results indicate that the proposed method provides effective technical support for refined monitoring and control in smart factory applications.

关键词

Keywords

references

Zhou J Z , Shi T , Qian Q M , et al . Protocol for the design and accelerated optimization of a waste-to-energy system using AI tools [J ] . STAR Protocols , 2023 , 4 ( 4 ): 102685 .

延炜 . 提高柴油收率的优化措施 [J ] . 炼油与化工 , 2021 ( 1 ): 29 - 31 .

Yan W . Optimization measures for improving diesel yield [J ] . Refining and Chemical Industry , 2021 ( 1 ): 29 - 31 .

聂红 , 张锐 , 陈文斌 , 等 . 柴油质量升级加氢催化剂与工艺技术的开发及应用 [J ] . 化学反应工程与工艺 , 2025 , 41 ( 5 ): 587 - 595 .

Nie H , Zhang R , Chen W B , et al . Development and application in upgrading diesel quality with hydrogenation catalysts and process technology [J ] . Chemical Reaction Engineering and Technology , 2025 , 41 ( 5 ): 587 - 595 .

Pang Z Y , Huang P , Lian C , et al . Data-driven prediction of product yields and control framework of hydrocracking unit [J ] . Chemical Engineering Science , 2024 , 283 : 119386 .

Zheng Q Q , Fan Y J , Zhou Z , et al . Research on product yield prediction and benefit of tuning diesel hydrogenation conversion device based on data-driven system [J ] . Energies , 2023 , 16 ( 14 ): 5332 .

Lim B , Arık S Ö , Loeff N , et al . Temporal Fusion Transformers for interpretable multi-horizon time series forecasting [J ] . International Journal of Forecasting , 2021 , 37 ( 4 ): 1748 - 1764 .

Gmeinbauer J , Ramakrishnan C , Reichhold A . Prediction of FCC product distributions by means of feed parameters [J ] . Oil Gas European Magazine , 2004 , 120 ( 3 ): 29 - 33 .

谢洋 , 陆新建 , 邵晓雯 , 等 . 基于MLP-RF组合模型预测加氢裂化装置柴油产量 [J ] . 化学工程 , 2024 , 52 ( 11 ): 66 - 70 .

Xie Y , Lu X J , Shao X W , et al . Prediction of productivity of diesel in hydrocracking unit based on MLP-RF combined model [J ] . Chemical Engineering , 2024 , 52 ( 11 ): 66 - 70 .

Yang F , Xu M , Lei W Q , et al . Artificial intelligence methods applied to catalytic cracking processes [J ] . Big Data Mining and Analytics , 2023 , 6 ( 3 ): 361 - 380 .

曹跃 , 余冲 , 李智 , 等 . 工业数据驱动的加氢裂化装置多工况切换过渡状态检测 [J ] . 化工学报 , 2023 ( 9 ): 3841 - 3854 .

Cao Y , Yu C , Li Z , et al . Industrial data driven transition state detection with multi-mode switching of a hydrocracking unit [J ] . CIESC Journal , 2023 ( 9 ): 3841 - 3854 .

Vieira C A , Victor dos Santos P , dos Santos Lima H O , et al . Application of machine learning models for predicting variables of industrial processes: a case study in a petrochemical industry [C ] // 2024 IEEE Latin American Conference on Computational Intelligence (LA-CCI) . November 13-15, 2024 , Bogota D.C., Colombia. IEEE , 2024 : 1 - 6 .

Hochreiter S , Schmidhuber J . Long short-term memory [J ] . Neural Computation , 1997 , 9 ( 8 ): 1735 - 1780 .

Zheng W D , Chen G . An accurate GRU-based power time-series prediction approach with selective state updating and stochastic optimization [J ] . IEEE Transactions on Cybernetics , 2022 , 52 ( 12 ): 13902 - 13914 .

Panjapornpon C , Bardeeniz S , Hussain M A . Improving energy efficiency prediction under aberrant measurement using deep compensation networks: a case study of petrochemical process [J ] . Energy , 2023 , 263 : 125837 .

王琳 , 张永渝 , 金涛 , 等 . 基于稀疏特征提取和门控循环单元的炼化装置能耗预测方法研究 [J ] . 现代化工 , 2025 , 45 ( 10 ): 253 - 258 .

Wang L , Zhang Y Y , Jin T , et al . Refinery energy consumption prediction method based on sparse feature extraction and gated recurrent unit [J ] . Modern Chemical Industry , 2025 , 45 ( 10 ): 253 - 258 .

Geng Z Q , Zhang Y H , Li C F , et al . Energy optimization and prediction modeling of petrochemical industries: an improved convolutional neural network based on cross-feature [J ] . Energy , 2020 , 194 : 116851 .

刘浩宇 , 王晨 , 倪元 , 等 . 信号分解与TCN-LSTM构建及在石化企业外排口水质预测中的应用 [J ] . 环境化学 , 2025 , 44 ( 12 ): 4809 - 4818 .

Liu H Y , Wang C , Ni Y , et al . Research on signal decomposition and the construction of TCN-LSTM applied to the prediction of wastewater quality at the discharge outlets of petrochemical plants [J ] . Environmental Chemistry , 2025 , 44 ( 12 ): 4809 - 4818 .

Vaswani A , Shazeer N , Parmar N , et al . Attention is all you need [C ] // Proc of the 31st International Conference on Neural Information Processing Systems . Red Hook, NY : Curran Associates Inc , 2017 : 6000 - 6010 .

杨晔 , 卢建刚 . 基于融合Transformer的门尼系数预测建模研究 [J ] . 化工学报 , 2025 , 76 ( 1 ): 266 - 282 .

Yang Y , Lu J G . Mooney viscosity prediction modeling based on fusion Transformer [J ] . CIESC Journal , 2025 , 76 ( 1 ): 266 - 282 .

张秦伟 , 张菁雯 , 郑斌 , 等 . 基于MIC-CCF-Informer的制氢转化炉出口温度预测 [J ] . 现代化工 , 2025 , 45 ( 12 ): 229 - 234 .

Zhang Q W , Zhang J W , Zheng B , et al . Prediction of temperature at outlet of hydrogen production reformer based on MIC-CCF-Informer network [J ] . Modern Chemical Industry , 2025 , 45 ( 12 ): 229 - 234 .

21 ] Maldonado-Cruz E , Pyrcz M J . Multi-horizon well performance forecasting with temporal fusion transformers [J ] . Results in Engineering , 2024 , 21 : 101776 .

Kraskov A , Stögbauer H , Grassberger P . Estimating mutual information [J ] . Physical Review E , 2004 , 69 ( 6 ): 066138 .

朱春梦 , 李增 , 柳楠 , 等 . 基于自编码器和多尺度符号转移熵的FCC沉降器跑剂故障检测 [J ] . 化工学报 , 2025 , 76 ( 9 ): 4512 - 4523 .

Zhu C M , Li Z , Liu N , et al . Fault detection of catalyst loss in FCC disengager based on autoencoder and multi-scale symbolic transfer entropy [J ] . CIESC Journal , 2025 , 76 ( 9 ): 4512 - 4523 .

Ji C , Ma F Y , Wang J H , et al . Real-time industrial process fault diagnosis based on time delayed mutual information analysis [J ] . Processes , 2021 , 9 ( 6 ): 1027 .

Li M , Xu X H , Hu L . Multivariate conditional Granger causality analysis based on feature selection and Backward Time-lagged Variable Selection algorithm [C ] // 2024 14th Asian Control Conference (ASCC) . July 5-8, 2024 , Dalian, China. IEEE , 2024 : 1879 - 1884 .

王雅琳 , 夏海兵 , 袁小锋 , 等 . 基于趋势相似度分析的多重时滞辨识及其在加氢裂化流程中的应用 [J ] . 化工学报 , 2018 , 69 ( 3 ): 1149 - 1157 .

Wang Y L , Xia H B , Yuan X F , et al . Multi-delay identification by trend-similarity analysis and its application to hydrocracking process [J ] . CIESC Journal , 2018 , 69 ( 3 ): 1149 - 1157 .

杨晨 , 彭小龙 , 朱苏阳 , 等 . 一种基于时序注意力动态卷积的油气井产量预测方法 [J ] . 油气藏评价与开发 , 2025 , 15 ( 6 ): 1046 - 1055 .

Yang C , Peng X L , Zhu S Y , et al . An oil and gas well production prediction method based on temporal attention and dynamic convolution [J ] . Petroleum Reservoir Evaluation and Development , 2025 , 15 ( 6 ): 1046 - 1055 .

He G , Luo L , Zhou L , et al . Deep learning prediction of yields of fluid catalytic cracking via differential evolutionary dual-stage attention-based LSTM [J ] . Fuel , 2024 , 370 : 131826 .

郭子越 , 金莹 , 朱孙泳 , 等 . GAST-Net：基于门控循环单元和注意力机制的发射光谱层析重建预测 [J ] . 光学学报 , 2025 , 45 ( 16 ): 198 - 211 .

Guo Z Y , Jin Y , Zhu S Y , et al . GAST-net: reconstruction and prediction by emission spectral tomography based on gated recurrent units and attention mechanism [J ] . Acta Optica Sinica , 2025 , 45 ( 16 ): 198 - 211 .

Song C R , Yang H D , Cai J Y , et al . Multi-energy load forecasting via hierarchical multi-task learning and spatiotemporal attention [J ] . Applied Energy , 2024 , 373 : 123788 .

Zhu Y Q , Yang F , Torgashov A . Causal-Transformer: Spatial-temporal causal attention-based transformer for time series prediction [J ] . IFAC-PapersOnLine , 2024 , 58 ( 14 ): 79 - 84 .

孙毅 , 邵远楫 , 单继宏 , 等 . 基于先验时滞特征的线缆捻制丝股张力均衡控制方法 [J ] . 计算机集成制造系统 , 2024 , 30 ( 9 ): 3174 - 3182 .

Sun Y , Shao Y J , Shan J H , et al . Strands tension balance control method of cable twisting based on characteristics of priori time-delay [J ] . Computer Integrated Manufacturing Systems , 2024 , 30 ( 9 ): 3174 - 3182 .

王笑辰 , 李世娟 , 黄文胜 , 等 . 融合先验知识与多源异构数据的临床误诊风险预测与特征识别研究 [J/OL ] . 数据分析与知识发现 , 2026 : 1 - 18 . ( 2026-01-02 )[ 2026-01-02 ] . https://link.cnki.net/urlid/10.1478.g2.20260113.1308.002 https://link.cnki.net/urlid/10.1478.g2.20260113.1308.002 .

Wang X C , Li S J , Huang Wensheng , et al . Clinical misdiagnosis risk prediction and feature identification based on prior knowledge and multi-source heterogeneous data [J/OL ] . Data Analysis and Knowledge Discovery , 2026 : 1 - 18 . ( 2026-01-02 )[ 2026-01-02 ] . https://link.cnki.net/urlid/10.1478.g2.20260113.1308.002 https://link.cnki.net/urlid/10.1478.g2.20260113.1308.002 .

张京新 , 何皎洁 , 蔡庆旺 , 等 . 基于二次分解和BiLSTM的污水厂出水COD浓度预测 [J ] . 化工学报 , 2025 , 76 ( 6 ): 2859 - 2871 .

Zhang J X , He J J , Cai Q W , et al . Prediction of COD concentration in wastewater treatment plant effluent based on secondary decomposition and BiLSTM [J ] . CIESC Journal , 2025 , 76 ( 6 ): 2859 - 2871 .

Chen T Q , Guestrin C . XGBoost: a scalable tree boosting system [C ] // Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . San Francisco California USA . ACM , 2016 : 785 - 794 .

Rumelhart D E , Hinton G E , Williams R J . Learning representations by back-propagating errors [J ] . Nature , 1986 , 323 ( 6088 ): 533 - 536 .

Zhou H Y , Zhang S H , Peng J Q , et al . Informer: beyond efficient transformer for long sequence time-series forecasting [J ] . Proceedings of the AAAI Conference on Artificial Intelligence , 2021 , 35 ( 12 ): 11106 - 11115 .

Fay M P , Brittain E H , Shih J H , et al . Causal estimands and confidence intervals associated with Wilcoxon-Mann-Whitney tests in randomized experiments [J ] . Statistics in Medicine , 2018 , 37 ( 20 ): 2923 - 2937 .

Shrikumar A , Greenside P , Kundaje A B . Learning important features through propagating activation differences [C ] // International Conference on Machine Learning ., 2017

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Soft sensor development based on deep extended variational autoencoder with just-in-time learning

Optimal method of selecting silicon content data in blast furnace hot metal based on k-means++

Soft sensor modeling for dynamic liquid level of oil well based on fuzzy inference adaptive updating

Effluent BOD soft measurement based on mutual information and self-organizing RBF neural network

Atmospheric tower soft sensor based on regression and mutual information of kernel slow features

Related Author

XIONG Shengjie

XIE Li

XU Liang

CAO Yuqing

YANG Huizhong

Linzi YIN

Yuyin GUAN

Zhaohui JIANG

Related Institution

School of Internet of Things Engineering, Jiangnan University

Wuxi Advance Technologies

College of Physics and Electronics, Central South University

School of Automation, Central South University

School of Electrical Engineering, Shenyang University of Technology

⁰