王振阳等:基于支持向量回归与极限学习机的高炉铁水温度预测 575· 预测值相较测量基准值的偏差波动;选取命中率 及误差波动方面表现较佳,对铁水温度±10℃范 (Hit rate,HR)用以表征在工艺可接受范围内的模 围内的预测命中率为94%. 型预测可信度,其计算方法分别如式(15)至(18) 所示: 参考文献 [1]Wang Z Y,Zhang J L,Liu Z J,et al.Status,technological MAE=-. (15) progress,and development directions of the ironmaking industry in China.Ironmaking Steelmaking,2019,46(10):937 [2] (16) Wang Z Y,Zhang J L,An G,et al.Analysis on the oversize blast furnace desulfurization and a sulfide capacity prediction model based on congregated electron phase.Metall Mater Trans B,2016, RMSE= 日2- (17) 47(1):127 [3]Martin R D,Obeso F,Mochon J,et al.Hot metal temperature prediction in blast furnace using advanced model based on fuzzy R=HR×10% logic tools.Ironmaking Steelmaking,2007,34(3):241 n [4] Jimenez J,Mochon J,de Ayala J S,et al.Blast furnace hot metal (18) 1,-y≤c temperature prediction through neural networks-based models./S// HR;= m,2004.44(3):573 0,-为>c [5] Ding Z Y,Zhang J,Liu Y.Ensemble non-Gaussian local 其中,n为测试集中样本总数量,和:分别为铁 regression for industrial silicon content prediction./SI/Int,2017, 水温度预测值与测量值,c为命中率判定边界值, 57(11):2022 本文选取c值为10℃,即以误差(-10℃,10℃)范 [6] Gao C H.Chen J M.Zeng J S,et al.A chaos-based iterated 围内为基准进行命中率计算,计算结果如表3所 multistep predictor for blast fumace ironmaking process.A/ChE 2009,55(4):947 示.由此可知,无论是从偏差绝对值与相对值角 [7]Cui G M,Li J,Zhang Y,et al.Prediction modeling study for blast 度,亦或是从偏差波动与命中率角度,SVR与ELM fumace hot metal temperature based on T-S fuzzy neural network 算法均对铁水日平均温度实现了较好的预测.相 model.Iron Steel,2013,48(11):11 较而言,SVR算法对铁水温度的预测优于ELM算 (崔桂梅,李静,张勇,等.基于TS模糊神经网络模型的高炉铁 法,其平均绝对误差、平均绝对百分比误差、均方 水温度预测建模.钢铁,2013,48(11):11) 根误差与±10℃以内的命中率表现较佳 [8] Cui G M,Cheng S.Predictive modeling of blast furnace temperature by using distributed neural network model.J Iron 表3SVR与ELM算法铁水温度预测结果综合定量表征 Steel Res,.2014,26(6):27 Table 3 Quantitative characterization of SVR and ELM model (崔桂梅,程史.基于分布式神经网路模型的高炉炉温预测建模 prediction results of hot metal temperature 钢铁研究学报,2014,26(6):27) Model MAPE/% MAE/℃ RMSE/C HP±10℃)M% [9] Shi L,Tang J J,Yu T,et al.Parameters prediction on furnace SVR 0.29 4.33 5.60 94.0 temperature on blast furnace based on a nonlinear spline ELM 0.31 4.69 6.09 88.5 transform-PLS.J Iron Steel Res,2013,25(2):20 (石琳,汤佳佳,于涛,等.基于样条变换的非线性PLS的反应高 炉炉温的参数预测.钢铁研究学报,2013,25(2):20) 3结论 [10]Zhang X M,Kano M,Matsuzaki S.A comparative study of deep and shallow predictive techniques for hot metal temperature (1)正相关影响铁水温度变化的特征参量主 prediction in blast furnace ironmaking.Compu Chem Eng,2019, 要为煤气利用率、铁水硅含量和利用系数等:负相 130:106575 关影响铁水温度变化的特征参量主要为炉腹煤气 [11]Zhang X M,Kano M,Matsuzaki S.Ensemble pattern trees for 指数、水温差和焦丁比等.铁水温度受到高炉直 predicting hot metal temperature in blast furnace.Comput Chem 接与间接还原度以及热量平衡相关特征参量的影 Eng,2019,121:442 [12]Diaz J,Fernandez F J,Prieto MM.Hot metal temperature 响较大 forecasting at steel plant using multivariate adaptive regression (2)基于SVR与ELM算法构建的模型均对铁 splines.Metals,2020,10(1):41 水温度实现了较好的预测效果,前者在预测值与 [13]Hashimoto Y,Sawa Y,Kano M.Online prediction of hot metal 误差值散点分布,平均绝对误差和百分比误差,以 temperature using transient model and moving horizon estimation.预测值相较测量基准值的偏差波动;选取命中率 (Hit rate,HR)用以表征在工艺可接受范围内的模 型预测可信度,其计算方法分别如式(15)至(18) 所示: MAE = 1 n · ∑n i=1 |yˆi −yi | (15) MAPE = 1 n · ∑n i=1 yˆi −yi yi ×100% (16) RMSE = tv 1 n · ∑n i=1 (yˆi −yi) 2 (17) HR = 1 n · ∑n i=1 HRi ×100% HRi = 1, byi −yi ⩽ c 0, byi −yi > c (18) 其中,n 为测试集中样本总数量, yˆi 和 yi 分别为铁 水温度预测值与测量值,c 为命中率判定边界值, 本文选取 c 值为 10 ℃,即以误差(−10 ℃,10 ℃)范 围内为基准进行命中率计算,计算结果如表 3 所 示. 由此可知,无论是从偏差绝对值与相对值角 度,亦或是从偏差波动与命中率角度,SVR 与 ELM 算法均对铁水日平均温度实现了较好的预测. 相 较而言,SVR 算法对铁水温度的预测优于 ELM 算 法,其平均绝对误差、平均绝对百分比误差、均方 根误差与±10 ℃ 以内的命中率表现较佳. 表 3 SVR 与 ELM 算法铁水温度预测结果综合定量表征 Table 3 Quantitative characterization of SVR and ELM model prediction results of hot metal temperature Model MAPE/% MAE /℃ RMSE/℃ HP(±10 ℃)/% SVR 0.29 4.33 5.60 94.0 ELM 0.31 4.69 6.09 88.5 3 结论 (1)正相关影响铁水温度变化的特征参量主 要为煤气利用率、铁水硅含量和利用系数等;负相 关影响铁水温度变化的特征参量主要为炉腹煤气 指数、水温差和焦丁比等. 铁水温度受到高炉直 接与间接还原度以及热量平衡相关特征参量的影 响较大. (2)基于 SVR 与 ELM 算法构建的模型均对铁 水温度实现了较好的预测效果,前者在预测值与 误差值散点分布,平均绝对误差和百分比误差,以 及误差波动方面表现较佳,对铁水温度±10 ℃ 范 围内的预测命中率为 94%. 参 考 文 献 Wang Z Y, Zhang J L, Liu Z J, et al. Status, technological progress, and development directions of the ironmaking industry in China. Ironmaking Steelmaking, 2019, 46(10): 937 [1] Wang Z Y, Zhang J L, An G, et al. Analysis on the oversize blast furnace desulfurization and a sulfide capacity prediction model based on congregated electron phase. Metall Mater Trans B, 2016, 47(1): 127 [2] Martin R D, Obeso F, Mochon J, et al. Hot metal temperature prediction in blast furnace using advanced model based on fuzzy logic tools. Ironmaking Steelmaking, 2007, 34(3): 241 [3] Jimenez J, Mochon J, de Ayala J S, et al. Blast furnace hot metal temperature prediction through neural networks-based models. ISIJ Int, 2004, 44(3): 573 [4] Ding Z Y, Zhang J, Liu Y. Ensemble non-Gaussian local regression for industrial silicon content prediction. ISIJ Int, 2017, 57(11): 2022 [5] Gao C H, Chen J M, Zeng J S, et al. A chaos-based iterated multistep predictor for blast furnace ironmaking process. AIChE J, 2009, 55(4): 947 [6] Cui G M, Li J, Zhang Y, et al. Prediction modeling study for blast furnace hot metal temperature based on T-S fuzzy neural network model. Iron Steel, 2013, 48(11): 11 (崔桂梅, 李静, 张勇, 等. 基于T-S模糊神经网络模型的高炉铁 水温度预测建模. 钢铁, 2013, 48(11):11) [7] Cui G M, Cheng S. Predictive modeling of blast furnace temperature by using distributed neural network model. J Iron Steel Res, 2014, 26(6): 27 (崔桂梅, 程史. 基于分布式神经网络模型的高炉炉温预测建模. 钢铁研究学报, 2014, 26(6):27) [8] Shi L, Tang J J, Yu T, et al. Parameters prediction on furnace temperature on blast furnace based on a nonlinear spline transform-PLS. J Iron Steel Res, 2013, 25(2): 20 (石琳, 汤佳佳, 于涛, 等. 基于样条变换的非线性PLS的反应高 炉炉温的参数预测. 钢铁研究学报, 2013, 25(2):20) [9] Zhang X M, Kano M, Matsuzaki S. A comparative study of deep and shallow predictive techniques for hot metal temperature prediction in blast furnace ironmaking. Comput Chem Eng, 2019, 130: 106575 [10] Zhang X M, Kano M, Matsuzaki S. Ensemble pattern trees for predicting hot metal temperature in blast furnace. Comput Chem Eng, 2019, 121: 442 [11] Diaz J, Fernandez F J, Prieto M M. Hot metal temperature forecasting at steel plant using multivariate adaptive regression splines. Metals, 2020, 10(1): 41 [12] Hashimoto Y, Sawa Y, Kano M. Online prediction of hot metal temperature using transient model and moving horizon estimation. [13] 王振阳等: 基于支持向量回归与极限学习机的高炉铁水温度预测 · 575 ·