一种基于神经网络的燃烧模型及其先验性检验

A-priori Analysis of a New Combustion Model Based on Artificial Neural Network Method

本研究采用神经网络方法,对准确的化学反应速率与反应标量之间的非线性关系进行建模,发展了一种基于人工神经网络(ANN)的燃烧模型.借鉴动态二阶矩模型建模思想,将标量分布的梯度纳入神经网络模型的输入集中,进一步发展出梯度输入人工神经网络(ANN-G)模型.基于一个预混火焰直接数值模拟数据库,对人工神经网络模型进行先验性研究,发现对于反应区域较薄的反应步,ANN模型与ANN-G模型都能准确计算化学反应速率.在泛化验证中,ANN-G模型比ANN模型表现更好.

A new artificial neural network(ANN)based combustion model has been developed by modelling the non-linear relationship between the accurate chemistry reaction rate and the known scalars based on the ANN method.With the Dynamic Second-order Moment Closure model for reference,the gradient of the scalar distribution is added as extra input variable,and the artificial neural network with gradient input(ANN-G)model is then developed.The two models have been validated in the a-priori analysis on the basis of the direct numerical simulation database of a premixed flame.It is observed that both the ANN model and the ANN-G model can predict the chemical reaction rate more accurately than laminar chemistry closure model,especially for the reaction steps in the thin reaction zone.In addition,the ANN-G model has better performance than the ANN model in the generalization validation.