Drum index and sieving index are key indicators reflecting sinter quality. To address the problem of low accuracy in traditional models caused by ignoring correlations between indicators in the prediction of sinter quality indicators (drum index and sieving index), a prediction model based on multi-task learning and long short-term memory (MTL-LSTM) was proposed. First, features strongly correlated with quality indicators were selected as inputs. Then, the long short-term memory (LSTM) was used as a shared layer to capture the coupling relationship between indicators via multi-task learning. This method realized information interaction through the shared layer, effectively solving the problem of low accuracy in independent predictions of traditional models. The results show that the MTL-LSTM model has higher accuracy than other combination models, providing a new direction for the high-quality production of sinter.
为克服传统单任务学习方法在特征间信息交互方面的局限性,本文设计了一种多任务学习策略.首先,对烧结现场采集的原始数据进行清洗和处理,确保输入变量在后续建模中的稳定性;随后采用最大互信息系数(maximal information coefficient,MIC)技术提取与转鼓指数及筛分指数紧密相关的输入特征,提升特征选择的科学性.在此基础上,结合烧结过程变量高度非线性且耦合特性显著的特点,引入长短期记忆网络(LSTM)作为建模核心,利用其在处理复杂动态序列数据方面的优势,提取特征的时序演化规律.为更有效捕捉2个质量指标间潜在的关系,采用多任务学习架构中的硬参数共享机制,构建统一的共享特征提取层,在保留各自预测输出的同时促进信息融合.该结构能够增强泛化能力并降低模型过拟合风险.最终基于真实烧结产线数据训练所构建的MTL-LSTM模型,与现有几种主流组合模型进行了对比实验.结果表明,该模型在精度和稳定性方面均优于对照组,有效弥补了传统模型在处理多指标任务时对指标间关联性考虑不足的缺陷,从而为提升烧结矿质量提供了一种可行的预测方法.
LiuS, LiuX J, LyuQ, et al. Comprehensive system based on a DNN and LSTM for predicting sinter composition[J]. Applied Soft Computing, 2020, 95: 106574.
[2]
XiaG L, WuZ X, LiuM Y, et al. Prediction interval estimation of sinter drum index based on light gradient boosting machine and kernel density estimation[J]. Ironmaking & Steelmaking, 2023, 50(8): 909-920.
[3]
HuangQ Y, LiuZ H, LiuZ C, et al. The strength prediction model of iron ore sinter based on an artificial neural network[J]. Ironmaking & Steelmaking, 2023, 50(2): 159-166.
[4]
LiuS, LyuQ, LiuX J, et al. Synthetically predicting the quality index of sinter using machine learning model[J]. Ironmaking & Steelmaking, 2020, 47(7): 828-836.
[5]
NiuL L, LiuZ J, ZhangJ L, et al. Prediction of sinter chemical composition based on ensemble learning algorithms[J]. Journal of Sustainable Metallurgy, 2023, 9(3): 1168-1179.
[6]
HuJ, WuM, ZhangP, et al. Prediction performance improvement via anomaly detection and correction of actual production data in iron ore sintering process[J]. IEEE Transactions on Industrial Informatics, 2020, 16(12): 7602-7612.
[7]
LiY F, ZhangQ W, ZhuY, et al. A model study on raw material chemical composition to predict sinter quality based on GA-RNN[J]. Computational Intelligence and Neuroscience, 2022, 2022(1): 3343427.
[8]
DuS, WuM, ChenX, et al. An intelligent control strategy for iron ore sintering ignition process based on the prediction of ignition temperature[J]. IEEE Transactions on Industrial Electronics, 2020, 67(2): 1233-1241.
[9]
WuZ Z, ZhangX M, ZhaoZ P, et al. Transfer learning aid the prediction of sintering densification[J]. Ceramics International, 2020, 46(16): 25200-25210.
[10]
DuS, WuM, ChenL F, et al. Operating mode recognition based on fluctuation interval prediction for iron ore sintering process[J]. IEEE/ASME Transactions on Mechatronics, 2020, 25(5): 2297-2308.
[11]
ShaoH J, YiZ M, ChenZ, et al. Application of artificial neural networks for prediction of sinter quality based on process parameters control[J]. Transactions of the Institute of Measurement and Control, 2020, 42(3): 422-429.
ZhangJun-hong, ShenFeng-man, XieAn-guo. The application of G-BP in FeO content prediction during sintering [J]. Journal of Northeastern University (Natural Science), 2002, 23(11): 1073-1075.
[14]
LiZ P, FanX H, ChenG, et al. Optimization of iron ore sintering process based on ELM model and multi-criteria evaluation[J]. Neural Computing and Applications, 2017, 28(8): 2247-2253.
[15]
ChenX X, ShiX H, LanT. A semi-supervised linear-nonlinear prediction system for tumbler strength of iron ore sintering process with imbalanced data in multiple working modes[J]. Control Engineering Practice, 2021, 110: 104766.
ZhangZhen, LiXin, LiuSong, et al. Predictive model of sinter drum index based on multi-category production status[J]. China Metallurgy, 2022, 32(1): 27-35.
LiuRan, ZhangZhi-feng, LiuXiao-jie, et al. Prediction of sinter drum index based on convolutional neural network and process theory[J]. Journal of Iron and Steel Research, 2023, 35(6): 651-658.
[20]
LiD C, HuangW T, ChenC C, et al. Employing box plots to build high-dimensional manufacturing models for new products in TFT-LCD plants[J]. Neurocomputing, 2014, 142: 73-85.
[21]
BaiL, YaoL N, KanhereS S, et al. Passenger demand forecasting with multi-task convolutional recurrent neural networks[C]//Advances in Knowledge Discovery and Data Mining. Cham: Springer, 2019: 29-42.