基于贝叶斯优化XGBoost的石灰窑气预测  

Bayesian optimization based XGBoost for lime kiln gas prediction

作　　者：温后珍[1,2] 栾仪广 孟碧霞 陈德斌[4] WEN HouZhen;LUAN YiGuang;MENG BiXia;CHEN DeBin(School of Mechanical Science and Engineering,Northeast Petroleum University,Daqing 163000;Jiangxi Ruoshui New Material Technology Co.Ltd.,Ji'an 331509;Shandong Yulong Petrochemical Industrial Park Development Co.,Yantai 265713;Daqing High-tech Industrial Development Zone Automotive and High-end Equipment Industry Promotion Center,Daqing 163000,China)

机构地区：[1]东北石油大学机械科学与工程学院,大庆163000 [2]江西若水新材料科技有限公司,吉安331509 [3]山东裕龙石化产业园发展有限公司,烟台265713 [4]大庆高新技术产业开发区汽车和高端装备产业促进中心,大庆163000

出　　处：《北京化工大学学报（自然科学版）》2024年第3期114-121,共8页Journal of Beijing University of Chemical Technology(Natural Science Edition)

基　　金：东北石油大学青年科学基金(HBHZX202004);黑龙江省自然科学基金(QC2016003)。

摘　　要：石灰窑是碳酸钙产业的关键生产设备,窑气中的CO_(2)是生产碳酸钙的原料,CO_(2)浓度直接影响碳酸钙产量,然而石灰窑气浓度依靠产品产出后采样化验得到,存在严重的滞后性,无法作为石灰窑在线工艺参数调整的依据。因此提出一种基于贝叶斯优化的eXtreme Gradient Boosting石灰窑气浓度预测模型BO-XGBoost,根据历史数据预测1 h后的窑气浓度,为生产工艺参数的调整提供依据。该方法首先对石灰窑传感器数据集中的缺失值、异常值进行剔除、插补,然后统一窑气浓度检测历史数据的时间尺度,构成石灰窑气监测数据集,在此基础上提出针对石灰窑气的BO-XGBoost模型。模型经训练后,采用实际生产数据进行测试,并与Light Gradient Boosting Machine(Light-GBM)模型、Category Boosting(Catboost)模型预测结果进行比较,结果表明,所提模型可以实现高维数据集的超参数快速优化,且预测模型有较好的精度,均方根误差(RMSE)达到0.70,平均绝对百分比误差(MAPE)达到2.03%。Lime kilns are a key component of the calcium carbonate industry.Since CO_(2)in kiln gas is the raw material for calcium carbonate production,the CO_(2)concentration directly affects calcium carbonate output.However,the concentration of lime kiln gas is obtained by sampling and testing after product output,which has a severe lag and cannot be used as a basis for online process parameter adjustment of the lime kiln.Therefore,a BO-XGBoost-based lime kiln gas concentration prediction model is proposed to predict the kiln gas concentration after 1 h based on historical data,which provides a basis for adjusting the production process parameters.The method first rejects and interpolates the missing and abnormal values in the data set for the lime kiln sensor,and then unifies the historical data for the measured kiln gas concentration on a time scale to form the lime kiln gas monitoring data set.A BO-XGBoost model for lime kiln gas was then proposed based on this data set.The model was trained and tested with actual production data,and compared with the prediction results of LightGBM and Catboost.The results show that the model can achieve rapid optimization of hyperparameters in high-dimensional data sets,and the prediction model has good accuracy.The maximum values of the RMSE and the MAPE were 0.70 and 2.03%respectively.Application of this model has important practical significance in the development of intelligent and precise control of lime kiln production in calcium corbonate enterprises.

关 键 词：石灰窑 石灰窑气 XGBoost模型 贝叶斯优化 

分 类 号：TQ115[化学工程—无机化工]