摘要
针对直接拟合得到的磨煤机一次风量黑箱软测量模型收敛性差、预测准确度较低的问题,在对风量测量进行机理特性分析的基础上,建立了一种以黑箱建模为主的结构逼近式混合软测量模型。黑箱模型部分采用最小二乘支持向量机(LSSVM)算法,并将机理特性融入到LSSVM模型中,经过输入变量类变换层的整合,将辅助变量中的压力信号转换为差压类、压力与温度的比值作为密度类、与一次风管路阻力相关的测点信号转换为风门风阻类,作为LSSVM模型的输入变量。应用电厂实际运行数据对一次风量进行预测,与直接拟合的预测效果相比,所建立的混合软测量模型收敛性好、预测精度高(引用误差波动<0.54%),可为电厂生产中控制合理的风煤配比关系提供准确度更高的一次风量预测值。
To deal with the difficulty that black-box soft sensor model of the coal mill primary air flow by direct fitting has poor convergence and low predictive accuracy, a kind of structure-approaching hybrid soft sensor model is proposed which is consisted of black-box model mainly based on the mechanism analysis of air flow measurement. The black box model part based on the least squares support vector machine （LSSYM） algorithm contains the characteristics of the mechanism. After integration of the transformation layer for input variables, the pressure signals in the auxiliary variables are transformed to differential pressure types, and the ratio between pressure and temperature is used as the density type. Measuring point signals associated with pipe resistance of primary air are transformed to air drag type as the input variables of the LSSVM model. Using the actual operating data of the power plant to predict the primary air flow, the forecast result shows that the established hybrid soft sensor model has good convergence and high predictive accuracy （ the quoted errors are less than 0.54% ） compared with direct fitting, which can provide more accurate predictive values of the primary air flow for the reasonable proportion of air and coal in the production of the power plant.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2016年第8期1913-1919,共7页
Chinese Journal of Scientific Instrument
关键词
一次风量
混合建模
最小二乘支持向量机
收敛性
风煤配比
primary air flow
hybrid modeling
least squares support vector machine （LSSVM）
convergence
air-coal ratio
