【摘要】 高炉铁水w (Si)是现场管理高炉的重点指标之一,但现场取样存在滞后性。因此铁水w (Si)预测对于帮助现场提前管控炉况具有重要意义。基于实际高炉数据,应用箱线图、多重插补法以及均值标准化进行治理,夯实数据基础。应用MIC协同PCA技术构建铁水w(Si)最优表征输入集合,包含50个w (Si)高关联度高炉参数,以及携带92.9%信息的5个高炉衍生特征。将GWO与XGBoost融合建模预测下1h铁水w(Si),GWO作为一种优化手段可使命中率(±0.05%误差范围)提升10.00个百分点。与其他模型相对比,GWO-XGBoost模型取得最优预测效果,平均绝对百分比误差为7.88%,均方根差为0.038,命中率最高达到85.86%,模型误差集中在0.07%以内,预测值与真实值吻合度较高,能够为现场提供可靠的铁水w(Si)未来趋势。更多还原

【Abstract】 The hot metal w(Si) in the blast furnace was one of the key indicators for on-site management,but there was a delay in on-site sampling.Therefore,predicting the hot metal w(Si) was of great significance for helping to control furnace conditions in advance.Based on actual blast furnace data,governance was implemented using boxplots,multiple imputation,and mean normalization to strengthen the data foundation.MIC combined with PCA technology was applied to construct the optimal input set for characterizing the hot metal w(Si),which included 50 highly correlated blast furnace parameters and 5 derived features carrying 92.9% of the information.The GWO-XGBoost fusion model was used to predict the hot metal w(Si) one hour ahead.As an optimization method,GWO improved the hit rate by 10.00 percentage points(±0.05% error range).Compared to other models,the GWO-XGBoost model achieved the best prediction results,with a mean absolute percentage error of 7.88%,an root mean square error of 0.038,and a maximum hit rate of 85.86%.The model error was concentrated within 0.07%,and the predicted values closely matched the actual values,providing reliable future trend predictions for on-site hot metal w(Si).更多还原