摘要
针对天然气燃气轮机污染物预测难度大的问题,该文基于数值模拟方法研究了燃烧室头部旋流数、分级面积比、分级轴向距离等典型因素对污染物生成的影响,在此基础上提出了基于神经网络的燃气轮机污染物预测模型。研究结果表明:头部旋流数、分级面积比增大会导致燃烧室内部最高温度升高,NOx排放增多,而CO排放无明显变化;所构建的神经网络预测模型预测结果与数值模拟结果吻合,其中预测NOx平均误差为4.58%,CO平均误差为0.97%,证实了神经网络模型预测燃气轮机污染物排放可行且准确。
[Objective]As pollutant emission is an important technical index of gas turbines,pollutant emission prediction has become one of the active research topics.However,the irregular strong turbulent combustion process in the combustion chamber of natural gas turbines causes chaotic pollutant generation,and the characteristics of low-emission combustion are extremely complex.The influence law of various geometric factors on pollutant generation characteristics is not clear.Moreover,the common pollutant prediction methods have certain limitations.For example,the numerical simulation method needs to be combined with a complex dynamic mechanism,resulting in a long calculation time.Therefore,this paper proposes to apply a neural network to the prediction of gas turbine pollutant emissions and develop a new method for the rapid and accurate prediction of pollutant emissions.[Methods]Computational fluid dynamics-based numerical simulation was used to study the influence of typical structural factors,such as the number of first-stage swirling flow,the number of second-stage swirling flow,and the fractional area ratio,on pollutant generation in the gas turbine combustion chamber,and to elucidate the variation trends of pollutant generation for different structures.The data were divided into a training set and a test set.Four structural parameters,namely the first-level swirl number,the second-level swirl number,the graded area ratio,and the graded axial distance of the combustion chamber head,were defined as input variables;the NOx and CO emissions at the combustion chamber outlet were defined as output variables for neural network training calculation;and then the radical basis function(RBF)neural network prediction model was established.The model structure was determined as 4-22-2.[Results]The results showed that for the studied coaxial graded combustor,the increase in the swirl number will lead to the increased and backward movement of the vortex core in the return zone,and the increase of the graded area ratio will lead to an increase in the equivalent ratio in the center of the return zone,which will increase the intensity of chemical reactions in the combustor,the maximum temperature,and the NOx emission.The CO emission in the combustion chamber was not sensitive to the typical structural parameters of the combustion chamber head,and the CO emission at the combustion chamber outlet exhibited little change with the variation of different structural parameters,such as swirl number,fractional area ratio,and fractional axial distance.The established combustion chamber emission RBF neural network prediction model could accurately and rapidly predict the combustion chamber outlet emission under different structural parameters.The maximum prediction error of NOx emission was 12.28%,and the average error was 4.58%;the maximum prediction error of CO emission was 2.75%,and the average error was 0.97%.[Conclusion]In this study,the characteristics of gas turbine pollutant generation are analyzed via numerical simulation,and the results prove that the neural network prediction model can effectively predict the characteristics of gas turbine pollution emission with good feasibility and high accuracy.
作者
孙继昊
宋颖
石云姣
赵宁波
郑洪涛
SUN Jihao;SONG Ying;SHI Yunjiao;ZHAO Ningbo;ZHENG Hongtao(School of Power and Energy Engineering,Harbin Engineering University,Harbin 150001,China)
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2023年第4期649-659,共11页
Journal of Tsinghua University(Science and Technology)
基金
国家科技重大专项项目(Y2019-I-0022-0021,2017-III-0006-0031)。
关键词
燃料与燃烧
排放预测
污染物生成特性
神经网络
fuel and combustion
forecast of emissions
pollutant generation characteristics
neural network
