基于TST-LSTM模型的烧结料层透气性预测  

Air Permeability Prediction of Sinter Layer Based on TSTLSTM Model

作　　者：刘梦园 吴朝霞 王金杨 閤光磊 LIU Meng-yuan;WU Zhao-xia;WANG Jin-yang;XIA Guang-lei(School of Control Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China.)

机构地区：[1]东北大学秦皇岛分校控制工程学院,河北秦皇岛066004

出　　处：《东北大学学报（自然科学版）》2024年第10期1379-1385,共7页Journal of Northeastern University(Natural Science)

基　　金：河北省教育厅科学技术研究项目(BJ2021099).

摘　　要：烧结过程中烧结料层透气性对烧结矿的质量影响较大,因此需建立模型准确预测烧结料层透气性.由于传统编码-译码模型不能够满足时间序列的依赖关系,提出一种模型时序转换与长短期记忆网络(time‑series transformer-long short‑term memory,TST-LSTM)模型.此模型对变换神经网络模型的译码器部分进行处理,结合LSTM模型的优势,对烧结料层透气性进行了实时预测.最终用预测模型与传统的反向传播神经网络(back propagation neural network,BPNN)模型、支持向量回归(support vector regression,SVR)模型和长短期记忆(long short‑term memory,LSTM)模型的仿真结果进行比较.结果表明,TST-LSTM模型预测性能较好且稳定.根据实际烧结过程进行仿真预测,验证了所提方法的有效性.In the sintering process,the air permeability of the sinter layer significantly impacts sinter quality.Therefore,it is essential to construct a model for accurately air permeability prediction of the sinter layer.Due to the inadequacy of traditional coding‑decoding models in handling time series dependencies,time‑series transformer-long short‑term memory network(TST-LSTM)model is proposed.This model leverages the decoding component of the transformer model and combines the advantages of LSTM to achieve realtime prediction of air permeability of the sinter layer.Comparative analysis with simulation results from traditional backpropagation neural network(BPNN),support vector regression(SVR),and long shortterm memory(LSTM)models demonstrates that TST-LSTM exhibits superior and more stable prediction performance.The proposed method is validated through simulation predictions based on actual sintering processes.

关 键 词：烧结料层 透气性 预测模型 注意力机制 神经网络 变换神经网络模型 

分 类 号：TF046.4[冶金工程—冶金物理化学]