基于深度学习的乙烯层流扩散火焰温度和碳烟体积分数分布重建

Reconstruction of Temperature and Soot Volume Fraction Distribution of Ethylene Laminar Diffusion Flame Based on Deep Learning

利用高光谱仪对乙烯层流扩散火焰进行测量,选用多层感知器神经网络预测温度和碳烟体积分数分布,评估了模型的预测和抗噪能力,讨论了不同高度和燃料流量的火焰中温度和碳烟体积分数的分布情况.结果表明,神经网络能较为准确地重建实验火焰的温度和碳烟体积分数,并具有较强的抗噪能力;随着火焰高度的增加,碳烟体积分数峰值从两翼移向中心区域,温度趋于平缓,整体平均大小先增加后减小;随着燃料流量的降低,相同归一化高度的温度升高而碳烟体积分数降低.

A hyperspectral imager was used to measure the ethylene laminar diffusion flame,the multi-layer perceptron(MLP)neural network was used to predict the distributions of temperature and soot volume fraction,the prediction and anti-noise ability of the MLP model were evaluated,and the distributions of temperature and soot volume fraction in flames of different heights and fuel flow rates were discussed.The results show that the neural network can more accurately reconstruct the temperature and soot volume fraction of the experimental flame and has a strong anti-noise ability.As the flame height increases,the peak soot volume fraction moves from the wings to the central region of the flame,the temperature tends to be flat,and the overall average size first increases and then decreases.As the fuel flow rate decreases,the temperature at the same normalized height increases while the soot volume fraction decreases.