Prediction of end-point temperature in electric arc furnace based on e-FCNN
LU Hongbin1, ZHU Hongchun1, JIANG Zhouhua1,2, LI Huabing1,2, YANG Ce1
1. School of Metallurgy, Northeastern University, Shenyang 110819, Liaoning, China; 2. State Key Laboratory of Rolling and Automation, Northeastern University, Shenyang 110819, Liaoning, China
Abstract：The development of the EAF steelmaking short process is an important strategic way to realize the green development of the iron and steel industry. The end-point control of EAF steelmaking determines the quality of tapping and smelting efficiency, especially the end-point temperature control. The establishment of the prediction model to achieve the EAF end-point temperature prediction in advance helps to adjust the smelting process in time and realize the fast and efficient tapping operation. The EAF end-point temperature prediction model is mainly divided into the mechanism model and the data-driven model. Data-driven modeling is the main research direction at present, but the modeling process usually relies on a large amount of historical data, and it is difficult to achieve accurate end-point temperature prediction under small sample data conditions. Therefore, tightly combined with the metallurgical mechanism, with artificial intelligence algorithms as the core, established a highly adaptive EAF end-point temperature prediction model. The input parameters of the model were obtained by metallurgical mechanism and Pearson data correlation analysis. Based on the FCNN algorithm, the early stopping strategy was introduced, the e-FCNN algorithm was proposed to prevent the overfitting phenomenon of the FCNN algorithm, and the end-point temperature prediction model of the EAF was established based on the e-FCNN algorithm. Simulation results show that the e-FCNN model end-point temperature prediction error is within ±5 ℃ with a hit rate of 93.33%. In addition, CART, RF, ε-SVR, and v-SVR models were developed using hyperparametric random grid search under the condition of small-sample historical data, and the results show that the accuracy of the e-FCNN model is significantly better than others. Using the e-FCNN model to continuously track the actual production of 30 heats, the hit rate reaches 96.7% when the prediction error is within ±6 ℃, which can effectively guide the production. In the future, further improvement of the combination of mechanism and data-driven is the development direction of EAF end-point temperature prediction models.
陆泓彬, 朱红春, 姜周华, 李花兵, 杨策. 基于e-FCNN的电弧炉终点温度预报[J]. 钢铁, 2024, 59(1): 49-57.
LU Hongbin, ZHU Hongchun, JIANG Zhouhua, LI Huabing, YANG Ce. Prediction of end-point temperature in electric arc furnace based on e-FCNN[J]. Iron and Steel, 2024, 59(1): 49-57.
[1] 姜周华, 姚聪林, 朱红春, 等. 电弧炉炼钢技术的发展趋势[J]. 钢铁,2020,55(7):1.(JIANG Z H, YAO C L, ZHU H C, et al. Application of increment artificial neural network model to prediction of end-point carbon, phosphorus and temperature for an 100 t EAF steelmaking[J]. Iron and Steel,2020,55(7):1.) [2] 李志慧, 杨凌志, 胡航, 等. 新钢基于多元数据采集的电弧炉炼钢过程能量监控系统[J]. 中国冶金, 2022, 32(11): 97.(LI Z H, YANG L Z, HU H, et al. Energy monitoring system of EAF steelmaking process based on multivariate data acquisition in Xinsteel[J]. China Metallurgy, 2022, 32(11):97.) [3] 徐阿帆,杨俊峰,王小平,等.100 t电弧炉能效优化工艺实践[J]. 中国冶金, 2023, 33(3): 109.( XU A F, YANG J F, WANG X P, et al. Energy efficiency optimization process practice of 100 t electric arc furnace[J]. China Metallurgy, 2023, 33(3): 109.) [4] 李进. 连续加料式电弧炉炼钢工艺数学模型[D]. 西安:西安建筑科技大学,1999.(LI J. Continuous Charging Type Electric Arc Furnace Steelmaking Process Mathematical Model[D]. Xi’an:Xi’an University of Architecture and Technology,1999.) [5] 李青. 炼钢电弧炉全程动态模型与仿真研究[D]. 上海:上海大学,2003.(LI Q. Dynamic Modeling and Simulation of the Entire Steelmaking Arc Furnace[D]. Shanghai:Shanghai University,2003.) [6] VAN ENDE M A. Development of an electric arc furnace simulation model using the effective equilibrium reaction zone (EERZ) approach[J]. Journal of the Minerals,2022,74(4):1610. [7] 刘锟,刘浏,何平,等. 增量神经网络模型预报100t电弧炉终点碳、磷和温度的应用[J]. 特殊钢,2004(3):401.(LIU K, LIU L, HE P, et al. Application of increment artificial neural network model to prediction of end-point carbon, phosphorus and temperature for an 100 t EAF steelmaking[J]. Special Steel,2004(3):40.) [8] 袁平,王福利,毛志忠. 基于G-SVM的电弧炉终点预报研究[J]. 钢铁研究学报,2006,18(10):7.(YUAN P,WANG F L,MAO Z Z. Endpoint prediction of EAF based on G-SVM[J]. Journal of Iron and Steel Research,2006,18(10):7.) [9] 姜静,李华德,孙铁,等. 基于混合遗传算法的电弧炉终点目标温度预报模型[J]. 特殊钢,2007,28(5):22.(JIANG J, LI H D, SUN T, et al. Predictive model for end aim temperature of arc furnace based on hybrid genetic algorithm[J]. Special Steel,2007,28(5):22.) [10] YUAN P,MAO Z Z,WANG F L. Endpoint prediction of EAF based on multiple support vector machines[J]. Journal of Iron and Steel Research International,2007,14(2):20. [11] MESA FERNÁNDEZ J M,CABAL V Á,MONTEQUIN V R,et al. Online estimation of electric arc furnace tap temperature by using fuzzy neural networks[J]. Engineering Applications of Artificial Intelligence,2008,21(7):1001. [12] 赵倩. 多模型融合技术在电弧炉钢水终点温度预报中的应用[D]. 沈阳:东北大学,2010.(ZHAO Q. Some Application of Multi-model Fusion Technology for Prediction of the EAF Endpoint Temperature[D]. Shenyang:Northeastern University, 2010.) [13] 袁平,王福利,毛志忠. 基于案例推理的电弧炉终点预报[J]. 东北大学学报(自然科学版),2011,32(12):1673. (YUAN P,WANG F L,MAO Z Z. CBR based endpoint prediction of EAF[J]. Journal of Northeastern University(Natural Science),2011,32(12):1673.) [14] 宋水根,刘花,曾繁林. 电弧炉炼钢全过程钢水碳质量分数动态预报模型[J]. 中国冶金,2013,23(12):25. (SONG S G, LIU H, ZENG F L. Dynamic forecast model of carbon content in molten steel in EAF steelmaking process[J]. China Metallurgy,2013,23(12):25.) [15] BLAŽIČ A,ŠKRJANC I,LOGAR V. Soft sensor of bath temperature in an electric arc furnace based on a data-driven Takagi-Sugeno fuzzy model[J]. Applied Soft Computing,2021,113(1):107949. [16] LI C,MAO Z. Generative adversarial network-based real-time temperature prediction model for heating stage of electric arc furnace[J]. Transactions of the Institute of Measurement and Control,2022,44(8):1669. [17] WANG M,GAO C,AI X,et al. Whale optimization end-point control model for 260 tons BOF steelmaking[J]. ISIJ International,2022,62(8):1684. [18] 贺东风,黄涵锐. 基于聚类思想的转炉终点碳含量预测方法[J]. 冶金能源,2022,41(6):29.(HE D F,HUANG H R. Prediction method of end point carbon content of converter based on clustering idea[J]. Energy for Metallurgical Industry,2022,41(6):29.) [19] YANG Q,ZHANG J,YI Z. Predicting molten steel endpoint temperature using a feature-weighted model optimized by mutual learning cuckoo search[J]. Applied Soft Computing,2019,83(1):105675. [20] FENG W,ZHU Q,ZHUANG J,et al. An expert recommendation algorithm based on Pearson correlation coefficient and FP-growth[J]. Cluster Computing,2019,22(3):7401. [21] XUE Y,WANG Y,LIANG J. A self-adaptive gradient descent search algorithm for fully-connected neural networks[J]. Neurocomputing,2022,478(1):70. [22] WANG K,DOU Y,SUN T,et al. An automatic learning rate decay strategy for stochastic gradient descent optimization methods in neural networks[J]. International Journal of Intelligent Systems,2022,37(10):7334. [23] SCABINI L F S,BRUNO O M. Structure and performance of fully connected neural networks:Emerging complex network properties[J]. Physica A:Statistical Mechanics and its Applications,2023,615(1):128585. [24] LEI X,FAN Y,LI K C,et al. High-precision linearized interpretation for fully connected neural network[J]. Applied Soft Computing,2021,109(1):107572. [25] CHU J,LIU X,ZHANG Z,et al. A novel method overcomeing overfitting of artificial neural network for accurate prediction:Application on thermophysical property of natural gas[J]. Case Studies in Thermal Engineering,2021,28(1):101406. [26] BEJANI M M,GHATEE M. A systematic review on overfitting control in shallow and deep neural networks[J]. Artificial Intelligence Review,2021,54(8):6391. [27] FORMENTIN S,KARIMI A. Enhancing statistical performance of data-driven controller tuning via-regularization[J]. Automatica,2014,50(5):1514. [28] BILGIC B,FAN A P,POLIMENI J R,et al. Fast quantitative susceptibility mapping with L1-regularization and automatic parameter selection:Fast QSM with L1-regularization[J]. Magnetic Resonance in Medicine,2014,72(5):1444. [29] KOIVU A,KAKKO J P,MÄNTYNIEMI S,et al. Quality of randomness and node dropout regularization for fitting neural networks[J]. Expert Systems with Applications,2022,207(1):117938. [30] XIE L,CHEN X,BI K,et al. Weight-sharing neural architecture search:A battle to shrink the optimization gap[J]. ACM Computing Surveys,2022,54(9):1. [31] WANG Q,MA Y,ZHAO K,et al. A comprehensive survey of loss functions in machine learning[J]. Annals of Data Science,2022,9(2):187. [32] MORALES-HERNÁNDEZ A,VAN NIEUWENHUYSE I,ROJAS GONZALEZ S. A survey on multi-objective hyperparameter optimization algorithms for machine learning[J]. Artificial Intelligence Review,2023,56(8):8043. [33] VALSECCHI C,CONSONNI V,TODESCHINI R,et al. Parsimonious optimization of multitask neural network hyperparameters[J]. Molecules,2021,26(23):7254. [34] CHITTURI S R,VERPOORT P C,LEE A A,et al. Perspective:new insights from loss function landscapes of neural networks[J]. Machine Learning:Science and Technology,2020,1(2):023002. [35] SHAO Y,DIETRICH F M,NETTELBLAD C,et al. Training algorithm matters for the performance of neural network potential:A case study of Adam and the Kalman filter optimizers[J]. The Journal of Chemical Physics,2021,155(20):204108. [36] SINGH D,SINGH B. Investigating the impact of data normalization on classification performance[J]. Applied Soft Computing,2020,97:105524. [37] RUTKOWSKI L,JAWORSKI M,PIETRUCZUK L,et al. The CART decision tree for mining data streams[J]. Information Sciences,2014,266(1):1. [38] SCHONLAU M,ZOU R Y. The random forest algorithm for statistical learning[J]. The Stata Journal:Promoting Communications on Statistics and Stata,2020,20(1):3. [39] KAVAKLIOGLU K. Modeling and prediction of Turkey's electricity consumption using support vector regression[J]. Applied Energy,2011,88(1):368. [40] YU H,LU J,ZHANG G. An online robust support vector regression for data streams[J]. IEEE Transactions on Knowledge and Data Engineering,2022,34(1):150. [41] CHEN W,XIE X,WANG J,et al. A comparative study of logistic model tree,random forest,and classification and regression tree models for spatial prediction of landslide susceptibility[J]. Catena,2017,151(1):147. [42] HAN T,JIANG D,ZHAO Q,et al. Comparison of random forest,artificial neural networks and support vector machine for intelligent diagnosis of rotating machinery[J]. Transactions of the Institute of Measurement and Control,2018,40(8):2681. [43] SAKIZADEH M,MIRZAEI R,GHORBANI H. Support vector machine and artificial neural network to model soil pollution:A case study in Semnan Province,Iran[J]. Neural Computing and Applications,2017,28(11):3229.
2024, Vol. 59 
