基于人工神经网络的超临界小火焰模型研究  

Application of artificial neural networks to supercritical flamelet model

作　　者：高正伟 金台 宋昌成 罗坤[1] 樊建人[1] GAO Zheng-wei;JIN Tai;SONG Chang-cheng;LUO Kun;FAN Jian-ren(State Key Laboratory of Clean Energy Utilization,Zhejiang University,Hangzhou 310027,China;School of Aeronautics and Astronautics,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江大学能源清洁利用国家重点实验室,浙江杭州310027 [2]浙江大学航空航天学院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2021年第10期1968-1977,共10页Journal of Zhejiang University：Engineering Science

基　　金：国家重点研发计划资助项目(2016YFB0600102).

摘　　要：为了解决超临界小火焰燃烧模型数据库过于庞大,导致计算机内存不足和取值性能下降的问题,提出使用人工神经网络(ANN)进行建库的超临界小火焰/过程变量模型FPV-ANN.在先验性分析及在超临界水热火焰的大涡模拟计算中发现,FPV-ANN方法在温度、组分和其他目标变量的分布与传统FPV方法得到的结果吻合,说明FPV-ANN方法的准确性与传统FPV方法一致.由于人工神经网络小火焰库大小只有传统库的1%,FPV-ANN方法在大规模并行计算中消耗更少的计算机内存.FPV-ANN方法的计算速度比传统FPV方法提升了30%.可以看出,提出的FPV-ANN方法具有更好的计算性能.Artificial neural networks(ANN)were utilized to build the library for the flamelet/progress variable(FPV)model and develop the FPV-ANN approach aiming at the problem that the enlarged lookup tables of the flamelet-based combustion model make the computer memory insufficient and slow down the interpolation process.Both the priori analysis and the large-eddy simulation of supercritical hydrothermal flames show that the distributions of temperature,species and other target variables obtained by FPV-ANN and classical FPV method achieve overall good agreement,verifying the accuracy of the FPV-ANN approach.Since the size of the ANN library is only 1%of the classical library,the use of FPV-ANN approach can produce a significant reduction in computer memory consumption during the large-scale parallel simulation.The computational speed of FPV-ANN approach is 30%faster than the classical FPV approach,which confirms that FPV-ANN approach has better computational performance.

关 键 词：小火焰模型 燃烧模拟 人工神经网络(ANN) 小火焰库建库方法 计算性能 

分 类 号：TK16[动力工程及工程热物理—热能工程]