烧结机煤气消耗量预测模型的性能比较研究

Comparison of performance of models for prediction gas consumption of sinter strand

李涛 ¹王盛民 ¹刘刚 ²王桂伟²

扫码查看

作者信息

1. 中唯炼焦技术国家工程研究中心有限责任公司
2. 日照检验认证有限公司
折叠

摘要

针对钢铁企业烧结机煤气消耗量预测精度较低的问题,研究建立差分自回归移动平均模型(ARIMA)、长短期记忆网络模型(LSTM)和极端梯度提升模型(XGBoost),用于预测烧结机的高炉煤气消耗量,利用钢铁联合企业的实际数据对比验证了预测模型性能.结果表明,XGBoost模型的预测精度高于ARIMA模型和LSTM模型.XGBoost模型的MAPE为3.45％,RMSE为703.53 m3/min,R2为99.91％,鲁棒性和泛化能力较强.此外,为了强化预测模型与烧结机不同运行状态间的联系,对烧结机不同运行状态的煤气消耗量进行预测.LSTM模型在烧结机正常生产状态表现出最好的预测效果,XGBoost模型则在烧结机减产和增产状态预测效果最佳.

Abstract

To solve the problem of low prediction accuracy of sinter strand gas consumption in iron and steel industry,autoregressive integrated moving average model(ARIMA),long short-term memory network model(LSTM)and extreme gradient boosting model(XGBoost)are established to predict the blast furnace gas consumption.The performance of the prediction models is verified by comparing the actual data of iron and steel industry.The results show that the prediction accuracy of XGBoost model is higher than ARIMA model and LSTM model.The mean absolute percentage error of XGBoost model is 3.45％,root mean square error is 703.53 m3/min,R2 is 99.91％,and the robustness and general-ization ability of XGBoost model are strong.In addition,in order to strengthen the connection between the prediction models and different operating states of sinter strand,the gas consumption of different operating states is predicted.LSTM model shows the best prediction effect in normal production state of sinter strand,while XGBoost model shows the best prediction effect in production reduction and pro-duction increase state.

关键词

钢铁企业/烧结机/煤气消耗量预测/数据模型/运行状态

Key words

iron and steel industry/sinter strand/gas consumption prediction/data model/operat-ing state

引用本文复制引用

出版年

2024

冶金能源

中钢集团鞍山热能研究院有限公司

冶金能源

CSTPCD北大核心

影响因子：0.319

ISSN：1001-1617

段落导航