基于Seq2Seq深度学习模型的焦炉煤气发生量预测方法研究

摘要/Abstract

摘要： 为实现钢铁生产过程中副产煤气的精准预测,构建了基于序列到序列的深度学习模型.通过编码器计算输入序列的隐状态得到隐状态矩阵,并通过解码器对其进行解码得到预测结果.根据灰色关联度分析关联度较高的输入参数,针对钢铁生产中煤气产量不稳定波动的特点,利用箱线图和hampel滤波对原始数据的极端异常点和突变点进行处理,对输入模型分别进行单步和多步预测.结果表明:单步预测时基于Seq2Seq结构的模型较单一模型预测性能有所提高,其中LSTM2GRU模型对峰谷值拟合表现最优;多步预测时LSTM2GRU模型可有效降低模型性能下降趋势,通过在2个数据集与LSTM2LSTM模型和GRU2GRU模型对比发现,LSTM2GRU模型均方根误差分别下降了5.3%、5.6%和9%、7.7%,平均绝对误差分别下降了7.3%、7%和9.7%、7.8%.因此,LSTM2GRU模型相比其他模型更适合长尺度时间序列的预测,在模型中引入GRU结构提高了预测精度,缩短了预测耗时.

关键词: 煤气预测, 神经网络, 深度学习, Seq2Seq模型, 灰色关联度

Abstract: To achieve accurate prediction of by-product gas in the steel production process, a sequence-to-sequence based deep learning model is constructed here. The hidden state matrix is obtained by calculating the hidden state of the input sequence through the encoder, and then the prediction result is obtained by decoding it through the decoder. Input parameters with high correlation are analyzed according to grey correlation. For the characteristics of unstable fluctuations in steel production, box line plots and hampel filtering are used to process extreme outliers and abrupt changes in the raw data, which are input to the model for single and multi-step prediction respectively. The results show that the prediction performance of the models based on the Seq2Seq structure is improved compared to their corresponding single models for single-step prediction, in which the LSTM2GRU model performs best for peak and valley fits. The LSTM2GRU model can effectively reduce the declining trend of model performance in multi-step prediction. Compared with the LSTM2LSTM model and the GRU2GRU model on two data sets, it is found that the root mean square error of the LSTM2GRU model has decreased by 5.3%, 5.6% and 9%, 7.7%, respectively. The mean absolute error of the LSTM2GRU decreased by 7.3%, 7% and 9.7%, 7.8%, respectively. Therefore, the LSTM2GRU model is more suitable for forecasting long-scale time series compared to other models. In addition, the introduction of the GRU structure into the model has improved prediction accuracy and reduced prediction time.

Key words: gas prediction, neural network, deep learning, Seq2Seq, grey relational analysis

中图分类号:

TE09

王文婷, 刘姝君, 张耀聪, 杜小泽, 许潼. 基于Seq2Seq深度学习模型的焦炉煤气发生量预测方法研究[J]. 兰州理工大学学报, 2023, 49(5): 50-58.

WANG Wen-ting, LIU Shu-jun, ZHANG Yao-cong, DU Xiao-ze, XU Tong. Prediction of coke oven gas generation based on Seq2Seq[J]. Journal of Lanzhou University of Technology, 2023, 49(5): 50-58.

参考文献

[1] JIN F,ZHAO J,HAN Z Y,et al.A joint scheduling method for multiple byproduct gases in steel industry [J].Control Engineering Practice,2018,80:174-184.
[2] WANG S M,WATADA J,PEDRYCZ W.Granular robust mean-CVaR feedstock flow planning for waste-to-energy systems under integrated uncertainty [J].IEEE Transactions on Cybernetics,2014,44(10):1846-1857.
[3] 岳有军,户彦飞,赵辉,等.基于小波分析的ARIMA与LSSVM组合的高炉煤气预测 [J].计算机测量与控制,2015,23(6):2128-2131.
[4] 熊文真,赵娜.基于ARIMA-MC模型钢铁企业高炉煤气发生量预测 [J].工业安全与环保,2016,42(10):100-102.
[5] 李红娟,熊文真.钢铁企业副产煤气预测及优化调度 [J].钢铁,2016,51(8):90-98.
[6] 马世龙,乌尼日其其格,李小平.大数据与深度学习综述 [J].智能系统学报,2016,11(6):728-742.
[7] 刘渺.钢铁企业主工序分厂煤气量预测方法研究 [D].长沙:中南大学,2006.
[8] XIAO D F,YANG C J,SONG Z H.The forecasting model of blast furnace gas output based on improved BP network [J].Journal of Zhejiang University (Engineering Science Edition),2012,46(11):2103-2108.
[9] 李志刚,纪月,任雄朝.基于LSTM与ARIMA组合模型的高炉煤气产生量预测 [J].铸造技术,2018,39(11):2456-2460.
[10] GREFF K,SRIVASTAVA R K,KOUTNIK J,et al.LSTM: A search space odyssey [J].IEEE Transactions on Neural Networks and Learning Systems,2016,28(10):2222-2232.
[11] SHAHID F,ZAMEER A,MUNEEB M.Predictions for COVID-19 with deep learning models of LSTM,GRU and Bi-LSTM [J].Chaos,Solitons and Fractals,2020,140: 110212.
[12] WISEMAN S,RUSH A M.Sequence-to-sequence learning as beam-search optimization [C]//Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing.[S.l.]:ArXiv E-Prints,2016:1296-1306.
[13] JIN X B,ZHENG W Z,KONG J L,et al.Deep-learning forecasting method for electric power load via attention-based encoder-decoder with bayesian optimization [J].Energies,2021,14(6):1-18.
[14] GONG G J,An X N,MAHATO N K,et al.Research on short-term load prediction based on Seq2Seq model [J].Energies,2019,12(16):1-18.
[15] XU J L,WANG K,LIN C,et al.FM-GRU: A time series prediction method for water quality based on Seq2Seq framework [J].Water,2021,13(8):1031.
[16] 王博文,王景升,王统一,等.基于长短时记忆网络的Encoder-Decoder多步交通流预测模型 [J].重庆大学学报,2021,44(11):71-80.
[17] ZHANG Z C,LI M,LIN X,et al.Multistep speed prediction on traffic networks: A deep learning approach considering spatio-temporal dependencies [J].Transportation Research Part C:Emerging Technologies,2019,105:297-322.
[18] CHEN H Q,GUAN M X,LI H.Air quality prediction based on integrated dual LSTM model [J].IEEE Access,2021,9: 93285-93297.
[19] KRISTIANI E,YANG C T,HUANG C Y,et al.PM2.5 forecasting using LSTM sequence to sequence model in Taichung city[M]//Information Science and Applications.[S.l.]:Springer,2020: 497-507.
[20] FANG Z,CRIMIER N,SCANU L,et al.Multi-zone indoor temperature prediction with LSTM-based sequence to sequence model [J].Energy and Buildings,2021,245:111053.
[21] HWANG S,JEON G,JEONG J,et al.A novel time series based Seq2Seq model for temperature prediction in firing furnace process [J].Procedia Computer Science,2019,155: 19-26.
[22] PETLURI N,AL-MASRI E.Web traffic prediction of wikipedia pages [C]//2018 IEEE International Conference on Big Data.[S.l.]:IEEE,2018: 5427-5429.
[23] HUO Y H,YAN Y,DU D,et al.Long-term span traffic prediction model based on STL decomposition and LSTM [C]//2019 20th Asia-Pacific Network Operations and Management Symposium (APNOMS).[S.l.]:IEEE,2019: 1-4.
[24] GAO Z H.Stock price prediction with ARIMA and deep learning models [C]//2021 IEEE 6th International Conference on Big Data Analytics (ICBDA).[S.l.]:IEEE,2021: 61-68.
[25] BENGIO Y,SIMARD P,FRASCONI P.Learning long-term dependencies with gradient descent is difficult [J].IEEE Transactions on Neural Networks,1994,5(2):157-166.
[26] GRAVES A,MOHAMED A R,HINTON G.Speech recognition with deep recurrent neural networks [C]//2013 IEEE International Conference on Acoustics,Speech and Signal Processing.[S.l.]:IEEE,2013.
[27] CHO K,VAN MERRIENBOER B,BAHDANAU D,et al.On the properties of neural machine translation: Encoder-decoder approaches [C]//Eighth Workshop on Syntax,Semantics and Structure in Statistical Translation.[S.l.]:ArXiv E-Prints,2014:103-111.
[28] 孙宇航,刘洋.利用GRU神经网络预测横波速度 [J].石油地球物理勘探,2020,55(3):484-492,503.
[29] SUTSKEVER I,VINYALS O,LE Q V.Sequence to sequence learning with neural networks [C]//Advances in Neural Information Processing Systems 27.[S.l.]:NIPS,2014:3104-3112.
[30] CHO K,VAN MERRIENBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation [C]//2014 Conference on Empirical Methods in Natural Language Processing.[S.l.]:ArXiv E-Prints,2014:1724-1734.
[31] MAQSOOD M,BUKHARI M,ALI Z,et al.A residual-learning-based multi-scale parallel-convolutions-assisted efficient CAD system for liver tumor detection [J].Mathematics,2021,9(10):1133.
[32] KIM C H,KIM M,SONG Y J.Sequence-to-sequence deep learning model for building energy consumption prediction with dynamic simulation modeling [J].Journal of Building Engineering,2021,43:102577.