数据驱动的水泥立磨系统出风口温度预测研究  

Data-driven prediction of air outlet temperature in cement vertical mill systems

作　　者：吕景祥 叶建辉 石洋 刘清涛[1] 马玉钦 张得洋 Lv Jingxiang;YE Jianhui;SHI Yang;LIU Qingtao;MA Yuqin;ZHANG Deyang(Key Laboratory of Road Construction Technology and Equipment of MOE,Chang an University,Xi an 710064,China;Sinoma(Tianjin)Powder Technology Equipment Co.,Ltd.,Tianjin 300400,China)

机构地区：[1]长安大学道路施工技术与装备教育部重点实验室,西安710064 [2]中材(天津)粉体技术装备有限公司,天津300400

出　　处：《安全与环境学报》2025年第4期1633-1642,共10页Journal of Safety and Environment

基　　金：国家自然科学基金项目(51705428)。

摘　　要：水泥生产立磨出风口温度是判断立磨运行状态是否安全稳定的关键参数,对该参数提前预测可以减少立磨振动,提高运行稳定性,增加产量,降低能耗及相关碳排放。水泥立磨系统具有多参数、大时滞和非线性等复杂特性。针对上述问题,提出了基于互相关延时分析优化的非线性自回归外部输入(Nonlinear AutoRegressive with eXogenous inputs,NARX)神经网络,并用于立磨出风口温度预测。首先,采用皮尔逊相关性分析从多个参数中确定影响立磨出风口温度的关键参数。同时,利用互相关延时分析进行时滞分析,解决大时滞问题。其次,通过优化的NARX神经网络,实现非线性工况下温度的精准预测。案例验证结果表明,所提出模型的拟合度达到了0.99967,均方误差为0.56483,预测精度达到了98.4%以上。预测模型结果可指导立磨操作人员及时控制立磨振动,提高水泥产量并降低能耗和碳排放。The aim of this study is to enhance the operational stability and control efficiency of the vertical roller mill,ultimately increasing productivity while reducing energy consumption and carbon emissions during the cement grinding process through accurate prediction of the mill s air outlet temperature.This paper presents a data-driven approach for predicting the air outlet temperature of the mill using a Nonlinear AutoRegressive with eXogenous inputs(NARX)neural network.First,Pearson correlation analysis was performed on data collected from the cement production process to identify parameters that are strongly correlated with the air outlet temperature of the mill.Based on the results of this analysis,nine key parameters were selected.Additionally,five parameters believed to influence the outlet temperature of the mill were selected based on expert experience and a thorough analysis of the cement production process.Subsequently,the time-delay relationship between the fourteen identified parameters and the mill s air outlet temperature was analyzed using the cross-correlation function.Building on the time-delay analysis,the input-output structure and time-delay order of the NARX neural network were optimized for temperature prediction,with the selected range for the time-delay order set between 21 and 31.Additionally,the relationship between prediction accuracy and lead time was analyzed,revealing that the mean square error of the model decreased as the lead time was shortened.The results indicated that the optimized NARX neural network achieved a fitting degree of 0.99967,a mean square error of 0.56483,and a prediction accuracy exceeding 98.4%.These results demonstrate superior performance compared to classical backpropagation neural networks and random forest methods.The advance prediction of the mill s outlet temperature can serve as a valuable reference for vibration control,enabling proactive adjustments to stabilize the mill s operational status.This approach not only enhances productivity but also reduces e

关 键 词：环境工程学 数据驱动 皮尔逊相关性分析 延时分析 非线性自回归外部输入神经网络 

分 类 号：X511[环境科学与工程—环境工程]