摘要
排烟含氧量是评价燃烧过程好坏和锅炉优化运行的重要指标,也是调节最佳风煤比的主要依据。针对工业锅炉氧量计使用受限的问题,提出一种混合的软测量方法。为了提高烟气含氧量的软测量预测精度,本文分析与烟气含氧量有关的锅炉运行变量,从中确定8个,并采用核主成分析法进行参数处理,整合冗余,降低维数。经处理后得到的6个主成分,其累计贡献率达95.522%,以此作为最小二乘支持向量机软测量模型的输入。在此基础上,通过划分网格来改进交叉实验法,进而优化最小二乘支持向量机的2个参数。经优化得到的误差参数γ和径向基核函数参数σ^2分别为90.3和239.6,模型具有较高的训练精度。最后对某循环流化床锅炉进行建模仿真,利用采集的数据,分别建立最小二乘支持向量机、核主成分分析的最小二乘支持向量机和BP神经网络3种模型。应用3种模型对烟气含氧量进行预测,并采用3个模型性能指标进行对比分析。结果表明,基于核主成分分析的最小二乘支持向量机的工业锅炉烟气含氧量模型,在小样本条件下学习更加有效,建模采样过程更快,预测精度更高。该模型有助于实现工业锅炉烟气含氧量在线软测量。
It is necessary to adjust the best ratio of coal and wind to keep boilers under the condition of a good combustion performance, which is vital to the actual operation of the boilers. Since utilizing apparatuses to test the oxygen-content is difficult, a mixed soft-sensing method was proposed. In order to improve the prediction accuracy of the soft-sensing of flue gas oxygen-content, eight boiler operation parameters that connected to oxygen-content were analyzed. The least square support vector machine with the kernel principal component analysis and improved crossover experimental methods were used to analyze the related parameters. After calculating the mathematical formulas, six principal components were extracted and the accumulative contribution rate of them was up to 95.522%. Then, the redundant samples were eliminated so as to decrease the emulation of input dimension. Using the improved crossover experimental method to optimize the parameters of the least squares support vector machine, error parameters and RBF kernel parameters were obtained. Thus, three soft-sensing models of flue gas oxygen-content were established by using collected data. Three kings of model were least square support vector machine, the least square support vector machine with the kernel principal component analysis and BP neural network. These models were applied to estimate the circulating fluidized bed boiler. Comparing the prediction values with the actual measurement data and utilizing three property indexes to analyze these models' performance, the results show that the established model of the least square support vector machine with the kernel principal component analysis has better predictability and reliability under the small sample conditions. This model contributes to realize the online soft-sensing of flue gas oxygen-content.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2012年第7期834-838,共5页
Computers and Applied Chemistry
基金
福建省科技重大专项(闽科技[2008]88号)
关键词
工业锅炉
烟气含氧量
软测量
核主成分分析
最小二乘支持向量机
industrial boiler, flue gas oxygen content, soft-sensing, kernel principal component analysis, the least square support vector machine