Prediction system for vanadium content of molten iron in BF based on improved wavelet-TCN
WANG Kun-ming1, LIU Xiao-jie1, LI Xin1, LI Hong-wei1, ZHANG Shu-hui1, CHEN Shu-jun2
1. College of Metallurgy and Energy, North China University of Technology, Tangshan 063210, Hebei, China; 2. Chengde Steel Group Co., Ltd., Chengde 067000, Hebei, China
摘要 铁水钒含量作为冶炼钒钛磁铁矿高炉的重要经济指标,对其进行准确预测将对高炉后续提钒增效具有重要生产意义。利用小波-TCN组合时序模型对具有非线性、波动大等特点的高炉铁水钒含量进行预测。首先利用小波变换将原时间序列数据分解成多个噪声段和单个趋势段,然后选用TCN模型对小波变换后的噪声段和趋势段分别进行预测,最后将结果重构得到最终的预测结果。对于选取小波变换层数较复杂的问题,利用赫斯特系数能够表征数据可预测性的特点,提出小波变换后的平均赫斯特系数(公式)用于降低模型建立过程中小波变换层数选取的复杂度,从而改进小波-TCN组合时序模型。结果表明,改进后的预测模型对单一变量预测高效且准确,相对非改进模型运算时间减少150％左右。对于赫斯特系数大于0.5的预测数据,利用改进小波-TCN组合时序模型对铁水钒含量进行预测,预测结果数据的R2达到0.967,均优于LSTM、LSTM with Attention和TCN单一预测模型的预测效果;对铁水硅、硫含量和铁水温度数据进行单变量预测,其R2分别为0.953、0.942和0.933。该预测模型可高效准确地对高炉铁水质量单变量进行预测,并可为高炉冶炼过程中所产生的其他波动较大数据的单变量准确、高效预测提供参考方案。基于预测模型进行预测系统功能应用开发,能使操高炉操作人员直观了解高炉出铁质量各参数状况,对高炉出铁质量数据进行提前掌握,促进高炉稳定顺行。
Abstract：Vanadium content of hot metal is an important economic index for BF smelting of vanadium-titanium magnetite ore, its accurate prediction will be of great significance to the subsequent vanadium extraction and efficiency of BF. Wavelet-TCN combined with time series model was used to predict the vanadium content in hot metal, which has the characteristics of non-linearity and large fluctuation. First of all, the original time series data were decomposed into multiple noise segments and a single trend segment by wavelet transform. Then the TCN model was selected to predict the noise segment and trend segment after wavelet transform, and the final prediction result was obtained by reconstructing. For the complex problem of selecting layers of wavelet transform, based on the characteristic that Hearst coefficients could represent the predictability of data, the average Hearst coefficients after wavelet transform were proposed to reduce the complexity of selecting layers of wavelet transform in the process of modeling, thus, the wavelet-TCN combined time series model could be improved. The results indicate that the improved prediction model is efficient and accurate for single variable prediction, and the operation time is about 150% less than that of the non-improved model. For the prediction data that the Hearst coefficient is greater than 0.5, R2 of predicted data for vanadium content in hot metal based on the improved wavelet-TCN combined time series model is 0.967, which is better than that of LSTM, LSTM with Attention and TCN single prediction model. For the single variable prediction data of silicon, sulfur content and temperature of hot metal, R2 is 0.953, 0.942 and 0.933, respectively. This prediction model can predict the single variable of BF hot metal quality efficiently and accurately, and provide a reference scheme for accurate and efficient single variable prediction of other bigger fluctuating data in BF smelting process. The application is developed based on the prediction system function in this prediction model, which could make the BF operators understand the parameters of iron quality intuitively to grasp the iron quality data in advance, and promote the stable and smooth operation of BF.
王坤明, 刘小杰, 李欣, 李红玮, 张淑会, 陈树军. 基于改进小波-TCN的高炉铁水钒含量预测系统[J]. 中国冶金, 2022, 32(12): 15-24.
WANG Kun-ming, LIU Xiao-jie, LI Xin, LI Hong-wei, ZHANG Shu-hui, CHEN Shu-jun. Prediction system for vanadium content of molten iron in BF based on improved wavelet-TCN[J]. China Metallurgy, 2022, 32(12): 15-24.
LI B, CHEN X F. Wavelet-based numerical analysis: A review and classification[J]. Finite Elements in Analysis and Design, 2014, 81: 14.
[11]
LI P, HUA P, GUI D, et al. A comparative analysis of artificial neural networks and wavelet hybrid approaches to long-term toxic heavy metal prediction[J]. Scientific Reports, 2020, 10(1):13439.
[12]
Alhnaitya B, Kollias S, Georgios L, et al. An autoencoder wavelet based deep neural network with attention mechanism for multi-step prediction of plant growth[J]. Information Sciences, 2021, 560:35.
[13]
Chen K, Laghrouche S, Djerdir A. Prognosis of fuel cell degradation under different applications using wavelet analysis and nonlinear autoregressive exogenous neural network[J]. Renewable Energy, 2021, 179:802.
[14]
Band S, Heggy E, Bateni S,et al. Groundwater level prediction in arid areas using wavelet analysis and Gaussian process regression[J]. Eegineering Applications of Computational Fluid Mechanics, 2021, 15(1):1147.
[15]
Sánchez-Granero M, Trinidad Segovia J, Pérez J. Some comments on Hurst exponent and the long memory processes on capital markets[J]. Physica A: Statistical Mechanics and its Applications, 2008, 387(22):5543.
[16]
Eom C, Choi s, Oh G, et al. Hurst exponent and prediction based on weak-form efficient market hypothesis of stock markets[J]. Physica A: Statistical Mechanics and its Applications, 2008, 387 (18):4630.
[17]
Hurst H. Long term storage capacity of reservoirs[J]. American Society of Civil Engineers, 1951, 116:509.
[18]
Qian B, Rasheed K. Hurst exponent and financial market predictability[J]. Proceedings of the Second IASTED International Conference on Financial Engineering and Applications, 2004, 12:1.
[19]
Doucouré B, Agbossou K, Cardenas A. Time series prediction using artificial wavelet neural network and multi-resolution analysis: Application to wind speed data[J]. Renewable Energy, 2016, 92:202.
Duan M. Time Series Predictability[D]. Milwaukee: Marquette University, 2002.
[22]
Bai S J, Kolter J, Koltun V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[J]. https://doi.org/10.48550/arXiv.1803.01271.
[23]
GONG L Y, YU M, JIANG S Y, et al. Deep learning based prediction on greenhouse crop yield combined TCN and RNN[J]. Sensors, 2021, 21(13):4537.
[24]
WANG Y W, DENG L, ZHENG L Y, et al. Temporal convolutional network with soft thresholding and attention mechanism for machinery prognostics[J]. Journal of Manufacturing Systems, 2021, 60:512.
[25]
CHANG P C, DANG J, DAI J R, et al. Real-time respiratory tumor motion prediction based on temporal convolutional neural network: Prediction model development study[J]. Journal of Medical Internet Research, 2021, 23(8):27235.
2022, Vol. 32 
