基于燃烧模拟和深度学习的钢包烘烤优化  

Ladle baking optimization based on combustion simulation and deep learning

作　　者：卢厚杨 张琦[1] 徐化岩 Lu Houyang;Zhang Qi;Xu Huayan(SEPA Key Laboratory on Eco-Industry,Northeastern University,Shenyang 110819,China;Beijing Yideyun Technology Co.,Ltd.,Beijing 100071,China)

机构地区：[1]东北大学国家环境保护生态工业重点实验室,沈阳110819 [2]北京易得云科技有限公司,北京100071

出　　处：《材料与冶金学报》2023年第3期211-217,共7页Journal of Materials and Metallurgy

基　　金：中央高校基本科研业务费项目(N2225047);国家自然科学基金项目(51874095)。

摘　　要：为了解决钢包烘烤的粗放式控制问题,提升钢包烘烤的燃烧效果,减少污染物排放量,提出了一种耦合燃烧模拟和深度学习技术进行数据驱动建模的钢包烘烤优化方法,同时采用了修改后的加权灰色气体模型确保CFD燃烧模拟模型计算在富氧条件下保持可靠.计算结果用于建立长短时记忆(LSTM)深度学习预测模型,预测氮氧化物(NO_(x))排放量、燃烧效率和升温速率,平均相对误差分别为0.278%,0.244%和0.189%.最后对计算结果进行多目标优化,可以在一定范围内减少氮氧化物排放量,提高燃烧效率和升温速率.结果表明:氮氧化物排放量、燃烧效率和升温速率的优化量分别为242.1 mg/m^(3),7.718%和3.631 K/h,达到了优化目的.In order to solve the extensive control problem of ladle baking,improve the combustion effect of ladle baking and reduce pollutant emissions,a ladle baking optimization method coupled with combustion simulation and deep learning techniques for data⁃driven modeling was proposed.A modified weighted gray gas model was adopted to ensure that CFD combustion simulation model calculations remain reliable under oxygen⁃rich conditions.The computational results were used to establish long-short term(LSTM)memory deep learning prediction models to predict NO_(x) emissions,combustion efficiency,and heat⁃up rate with average relative errors of 0.278%,0.244%,and 0.189%,respectively.Finally,the multi⁃objective optimization of the calculation results could reduce the NOx emission within a certain range,and improve combustion efficiency and heating rate.The results indicated that the optimized amounts of NO_(x) emission,combustion efficiency,and heating rate were 242.1 mg/m^(3),7.718%,and 3.631 K/h,respectively,which achieved the optimization purpose.

关 键 词：钢包烘烤 燃烧模拟 深度学习预测 多目标优化 

分 类 号：TF341[冶金工程—冶金机械及自动化]