基于图像处理的灰渣含碳量在线测量模型研究

Online prediction model of carbon content in ashusing image process technology

目的:针对燃煤锅炉飞灰及炉渣(以下简称灰渣)含碳量难以在线测量的现状,研究采用神经网络结合数字图像处理技术对灰渣含碳量进行实时测量的方法和系统的组成。方法:构建一套图像采集系统用来实现实时灰渣图像的获取及清晰、稳定的传输,在计算机Windows操作系统上利用OpenCV视觉库对图像进行灰度化、阈值分割、中值滤波、图像通道分离、合并等操作,获取灰渣图像在RGB、YUV、HSI三个颜色空间下最佳颜色特征参数,分析各图像颜色特征值与灰渣含碳量之间的对应关系,在此基础上建立基于粒子群算法优化BP神经网络的灰渣含碳量测量模型。结果:通过测量模型获取的灰渣含碳量实时数据与传统工业检测结果相比较,最大相对误差为8.10%,平均相对误差为5.66%,通过QT5(模型开发所用软件)时间测算函数的数据,在线检测模型及系统运行时间小于1 min。结论:建立的基于PSO-BP神经网络的灰渣含碳量测量模型,可在极短时间内完成图像采集、处理,从而可基本实现灰中未燃尽碳含量的在线测量,使运行人员准确了解锅炉实时的燃烧效率,对指导锅炉高效运行和节能降耗具有重大意义。

Aims:This paper aims to measure carbon content in the fly ash and slag of coal-fired boilers in real time.Methods:We proposed a novel method and built a corresponding system based on the neural network and digital image processing technology.An image acquisition system was built to acquire and transmit ash images in a stable and clear fashion.The OpenCV visual library based on the Windows operating system was applied to perform image graying,threshold segmentation,median filtering,image channel separation,merging and other operations.Based on this platform,we obtained the optimized color feature parameters of ash images in RGB,YUV and HSI color space and analyzed the corresponding relationship between image color eigenvalues and ash carbon content.An ash carbon content measurement model based on the particle swarm optimization BP neural network was established.Results:Compared with the traditional industrial test results,the maximum relative error of the method was 8.10%;and the average relative error was 5.66%.The elapsed time tested by QT5(software used for model development)was less than one minute.Conclusions:The established model based on the PSO-BP neural network can accurately measure the unburned carbon content in ash in real-time,which provides the real-time feedback on boiler combustion efficiency.This system is of vital significance for maintaining boilers in high efficiency and energy saving.