摘要
建立了一套可以识别和预测不同燃烧状态的实验系统。该系统维持空气流量稳定,通过调节液化石油气流量,得到稳定燃烧、脱火和扩散燃烧等典型的燃烧状态。利用视频采集卡在线采集火焰图像,图像识别程序分析出火焰的位置形状以及强度等特征信息,然后采用附加动量项与可变学习率的误差反向传播训练算法(BP)网络根据这些特征信息进行燃烧状态识别,最后预测程序根据识别结果对下一时刻的燃烧状态进行预测。实验结果表明:该系统可以准确地在线识别火焰燃烧状态,并能预测随后的火焰燃烧状态。
An experimental system is established to recognize and to predict the different combustion states. The flux of air is steady, the flux of the liquefied petroleum gas （LPG） is changed continuously by the system to get some typical combustion states, such as the steady combustion state, the flameout state and the diffusion combustion state. The combustion flame images are collected on line by the video capture card, and its characteristic parameters such as the flame＇s shape and position, and the radiation intensity are analyzed by the image recognition program. Based on these parameters, the combustion state is recognized by the back-propagation （BP） neural network with momentum and variable learning rate. The later combustion states are predicted by the prediction program. Experimental results show that the system can accurately recognize the different combustion states and predict the later combustion state on line.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2008年第4期468-471,共4页
Journal of Nanjing University of Science and Technology
