基于长短时记忆神经网络的带钢酸洗浓度预测
|
摘要
热轧带钢生产中一般通过酸洗来清除氧化铁皮。控制酸浓度在合理的范围内对于保证酸洗效果起着至关重要的作用,准确测量酸浓度才能更精准地对其进行控制。在机器学习、神经网络兴起的背景下,提出了基于长短时记忆(LSTM)神经网络的酸浓度预测方法。以某钢厂酸洗数据为对象,建立了LSTM酸浓度预测模型并对样本数据进行预测。试验结果表明,盐酸浓度、铁离子质量浓度的预测平均绝对误差分别为3.17 g/L,3.52 g/L,符合行业的误差规范要求;所提出的模型具备较高的预测精度,拥有较好的浓度预测性能,与传统统计模型相比,拥有更好的预测精度和更好的适用性。
In hot rolled strip production,iron oxide scale is usually removed by pickling. Controlling acid concentration in a reasonable range plays a vital role in ensuring the pickling effect. Accurate measurement of acid concentration can control it more accurately. Under the background of machine learning and the rise of neural network,an acid concentration prediction method based on long short-term memory(LSTM) is proposed. Based on the pickling data of a cold rolling mill in a steel mill,the LSTM acid concentration prediction model is established and the sample data is predicted. The results show that the average absolute errors of hydrochloric acid concentration and iron ion concentration are 3.17 g/L and 3.52 g/L respectively,which meet the requirements of industry error standards. The proposed model has higher prediction accuracy and better concentration prediction performance. Compared with the traditional statistical model,it has better prediction accuracy and better applicability.
In hot rolled strip production,iron oxide scale is usually removed by pickling. Controlling acid concentration in a reasonable range plays a vital role in ensuring the pickling effect. Accurate measurement of acid concentration can control it more accurately. Under the background of machine learning and the rise of neural network,an acid concentration prediction method based on long short-term memory(LSTM) is proposed. Based on the pickling data of a cold rolling mill in a steel mill,the LSTM acid concentration prediction model is established and the sample data is predicted. The results show that the average absolute errors of hydrochloric acid concentration and iron ion concentration are 3.17 g/L and 3.52 g/L respectively,which meet the requirements of industry error standards. The proposed model has higher prediction accuracy and better concentration prediction performance. Compared with the traditional statistical model,it has better prediction accuracy and better applicability.
引文
[1]余伟,王俊,刘涛.热轧钢材氧化及表面质量控制技术的发展及应用[J].轧钢,2017,34(3):1-6.
[2]He F,Li M,Wang B J.Multi-mode acid concentration prediction models of cold-rolled strip steel pickling process[J].Journal of Process Control,2014,24(6):916-923.
[3]郭建峰,李玉,安东.基于LM遗传神经网络的短期股价预测[J].计算机技术与发展,2017,27(1):152-155.
[4]张宇航,邱才明,贺兴,等.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-25.
[5]张春露,白艳萍.基于TensorFlow的LSTM模型在太原空气质量AQI指数预测中的应用[J].重庆理工大学学报:自然科学版,2018,32(8):137-141.
[6]Graves A.Generating sequences with recurrent neural networks[J].Computer Science,2013.
[7]Williams W,Prasad N,Mrva D,et al.Scaling recurrent neural network language models[C]//2015 IEEE International Conference on Acoustics,Speech and Signal Processing(ICASSP).IEEE,2015:5391-5395.
[8]Shalev-Shwartz S,Shamir O,Shammah S.Failures of gradientbased deep learning[DB/OL].(2017-04-26).https://arxiv.org/pdf/1703.07950v2.pdf.
[9]Pascanu R,Mikolov T,Bengio Y.On the difficulty of training recurrent neural networks[C]//Proceedings of the 30th International Conference on Machine Learning(ICML-13).2013:1310-1318.
[10]Graves A,Mohamed A R,Hinton G.Speech recognition with deep recurrent neural networks[C]//IEEE International Conference on Acoustics,Speech and Signal Processing.IEEE,2013.
[11]Srivastava N,Hinton G,Krizhevsky A,et al.Dropout:A simple way to prevent neural networks from overfitting[J].The Journal Of Machine Learning Research,2014,15(1):1929-1958.
[12]Kingmad D P,BA J Y.Adam:A method for stochastic optimization[C]//ICLR 2015,2015:1-15.
[2]He F,Li M,Wang B J.Multi-mode acid concentration prediction models of cold-rolled strip steel pickling process[J].Journal of Process Control,2014,24(6):916-923.
[3]郭建峰,李玉,安东.基于LM遗传神经网络的短期股价预测[J].计算机技术与发展,2017,27(1):152-155.
[4]张宇航,邱才明,贺兴,等.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-25.
[5]张春露,白艳萍.基于TensorFlow的LSTM模型在太原空气质量AQI指数预测中的应用[J].重庆理工大学学报:自然科学版,2018,32(8):137-141.
[6]Graves A.Generating sequences with recurrent neural networks[J].Computer Science,2013.
[7]Williams W,Prasad N,Mrva D,et al.Scaling recurrent neural network language models[C]//2015 IEEE International Conference on Acoustics,Speech and Signal Processing(ICASSP).IEEE,2015:5391-5395.
[8]Shalev-Shwartz S,Shamir O,Shammah S.Failures of gradientbased deep learning[DB/OL].(2017-04-26).https://arxiv.org/pdf/1703.07950v2.pdf.
[9]Pascanu R,Mikolov T,Bengio Y.On the difficulty of training recurrent neural networks[C]//Proceedings of the 30th International Conference on Machine Learning(ICML-13).2013:1310-1318.
[10]Graves A,Mohamed A R,Hinton G.Speech recognition with deep recurrent neural networks[C]//IEEE International Conference on Acoustics,Speech and Signal Processing.IEEE,2013.
[11]Srivastava N,Hinton G,Krizhevsky A,et al.Dropout:A simple way to prevent neural networks from overfitting[J].The Journal Of Machine Learning Research,2014,15(1):1929-1958.
[12]Kingmad D P,BA J Y.Adam:A method for stochastic optimization[C]//ICLR 2015,2015:1-15.