基于POD正交分解和BPNN神经网络的钢包底吹精炼快速预测模型  

Rapid prediction model of ladle bottom blowing refining based on POD decomposition and BP neural network

作　　者：聂忠赋 代威 贾琪 贺铸[1,2] 李光强[1,2] 王强[1,2] NIE Zhongfu;DAI Wei;JIA Qi;HE Zhu;LI Guangqiang;WANG Qiang(The State Key Laboratory of Refractories and Metallurgy,Wuhan University of Science and Technology,Wuhan 430o8l,Hubei,China;Key Laboratory for Ferrous Metallurgy and Resource Utilization of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430o8l,Hubei,China)

机构地区：[1]武汉科技大学省部共建耐火材料与冶金国家重点实验室,湖北武汉430081 [2]武汉科技大学钢铁冶金及资源利用省部共建教育部重点实验室,湖北武汉430081

出　　处：《钢铁研究学报》2025年第3期335-345,共11页Journal of Iron and Steel Research

基　　金：国家自然科学基金区域创新发展联合基金资助项目(U22A20173);湖北省自然科学基金一般面上资助项目(2023AFB796)。

摘　　要：为满足实时获取钢包底吹精炼三维流场数据的需求,建立了一种基于计算流体力学(Computational Fluid Dynamics,CFD)和本征正交分解(Proper Orthogonal Decomposition,POD)的钢包底吹三维流场快速预测模型。通过建立钢包底吹数值模型和水模型对其进行了模拟计算和实验,并建立了钢包单底吹的三维流场数据集,采用POD方法对数据集进行了模态与模态系数提取,通过反向传播神经网络(Back Propagation Neural Network,BPNN)构建了工况参数与模态系数间的映射关系,实现了对未来时刻钢包底吹内速度场和三相体积分数的快速预测。结果表明:所建立模型可以通过少量模态重构出钢包内流场的主要特征。POD-BPNN预测模型的计算结果准确率较高,计算结果平均相对误差均在4%以内。模型计算速度快,获取钢包内流场的平均计算时间由CFD全阶模拟所需的约246h减小至快速预测模型所需的约54.6h。In order to meet the requirement of real-time acquisition of 3D flow field data in ladle bottom-blowing refining,a fast prediction model of 3D flow field based on computational fluid dynamics(CFD)and proper orthogonal decomposition(POD)was established.The three-dimensional flow field data of single-nozzle bottom blowing of ladle were simulated and calculated by establishing the numerical model and the water model,and the data set was established.The mode and mode coefficient of the data set were extracted by the POD method.Through the back propagation neural network(BPNN),the mapping relationship between operating parameters and modal coefficients was constructed,and the velocity field and three-phase volume fraction in the bottomblowing ladle can be predicted quickly.The results show that the proposed model can reconstruct the main characteristics of the flow field in the ladle through a few modes.The POD-BPNN prediction model has a high accuracy of calculation results,and the average relative error of calculation results is less than 4%.The calculation speed of the model is fast,and the average calculation time to obtain the flow field in the ladle is reduced from about 246 h required by CFD simulation to about 54.6 h by the POD method.

关 键 词：本征正交分解 钢包底吹 渣眼 反向传播神经网络 计算流体力学 

分 类 号：TF769[冶金工程—钢铁冶金]