【作者】 刘晓阳；

【导师】 于洋；

【作者基本信息】 沈阳理工大学 ， 检测技术与自动化装置， 2009， 硕士

【摘要】 大型锅炉炉内煤粉的燃烧是一个发生在大空间范围、不断脉动的、具有明显三维特征的物理化学过程,因此燃烧工况的监测对于锅炉安全、经济效益、洁净运行具有重要的意义。针对大型锅炉内燃烧火焰变化的多样性,为能实时掌握炉内火焰燃烧状况,炉内三维温度场重建技术的研究则成为解决问题的关键。基于图像处理技术求解炉内燃烧火焰的三维温度分布是非接触的测量方式,是一种现代高科技控制手段。BP神经网络的研究和对三维温度场重建的仿真模拟及实验验证,则使三维温度场重建技术的应用更加具有现实意义。本文首先简要介绍利用比色法和BP神经网络摄取火焰图片温度值的基本方法。对BP神经网络模型的研究进行了深入的探讨,提出一种改进的BP神经网络方法,以此作为求取三维温度分布的手段,其具有传统优化求解法所不具备的很多优点。其次,本文对样本的选取、网络层数的确定以及学习步长等参数的最佳取值,都是采用边训练边修改的方式。最后,用训练好的BP网络进行了一系列仿真模拟和分析,并且通过热电偶点测量值对网络进行了实验验证,验证结果比较理想,同时也证明将BP神经网络用于三维温度分布的求取是可行的。更多还原

【Abstract】 The combustion of pulverized-coal in large-scale boiler is a complexly physical and chemical reaction, which is occurred in large space, fluctuating frequently, with conspicuous three-dimension character. Monitoring the combustion status is very important for boiler’s safety, economic benefits, clean operation. Because of the variety of burning flame in furnace of large-scale boiler, in order to know and control the condition of the burning flame in furnace, it is the key to research on how to reconstruct the three-dimensional temperature field in furnace.It is a non-contact measurement based on image processing technology to research on the three-dimensional temperature distribution of burning flame in furnace, which is modern high-tech control means. The investigation on BP neural networks and the simulation on reconstruction of three-dimensional temperature field make reconstructio -n technology more practical.In the paper, firstly, the basic method getting flame image temperature values is introduced in use of colorimetry and BP neural networks. BP neural networks model is delved into in depth and an improved BP neural networks is given, by which three-dimensional temperature distribution could be gotten, and which is superior to the traditional optimizing method. Then, the method selecting sample, network layer, the optimum values of study step and other parameters is to train the network and to modify it. Finally, a series of simulation and analysis with the trained network are done and the network is validated by the experiment of measuring temperature point to point with thermocouple. The result is ideal in degree. It is feasible to use BP neural networks to get three-dimensional temperature distribution.更多还原