电站锅炉主要热工过程参数软测量技术研究进展

Research status of soft measurement technology of typical thermal parameters for utility boilers

电站锅炉某些主要热工过程参数难以实现在线实时测量,从而制约了机组的高效、经济运行。为此,本文解析了基于统计分析的主元分析法和偏最小二乘法、基于人工智能的人工神经网络(ANN)法、基于统计学习理论的支持向量机法以及模糊理论法的建模方法。并以球磨机负荷、煤质、风煤比、烟气含氧量、飞灰含碳量、汽包水位、主蒸汽温度、省煤器积灰、污染物排放量等参数为对象,综述了各种软测量技术的研究现状。结果显示:对于飞灰含碳量等呈非线性特征的变量,基于核主元分析(KPCA)法建立其软测量模型,效果较好;当各变量的线性关联度高时,采用偏最小二乘回归(PLSR)法建立其软测量模型更为有效;对于人工神经网络法,当实际样本空间超出训练样本空间区域时,模型输出误差较大,因此实际工业过程中需定时对基于ANN法建立的模型参数进行校正;对于支持向量机(SVM)法还无成熟的指导方法,基于经验数据则对模型精度的影响较大,最小二乘支持向量机(LS-SVM)法的建模方法与传统SVM法相比,训练时间更短,结果更具确定性,更适合工业在线建模;模糊理论法不需要被测对象的精确数学模型,但模糊系统本身不具有学习功能,如果能够将其与ANN法等人工智能方法相结合,则可提高软测量的性能。因此,软测量技术的引入,使得难以在线测量的热工过程参数监测成为可能。

The difficulty in online measurement of some thermal parameters in utility boilers restricts the boilers’ efficient and economic operation.In this paper,the major models of soft measurement used in this field are firstly introduced briefly,including the method based on statistical analysis（like the principle component analysis（PCA）and partial least squares（PLS））,the artificial intelligence based artificial neural networks（ANN）,and the methods based on statistic studying theory（like the support vector machine（SVM）and fuzzy theory）.Afterwards,various thermal parameters are treated as the research objects of soft measurement to summarize the modeling and simulation of soft measurement technology.The parameters include the coal characteristics,load and air flow of ball mill,air-coal ratio,oxygen content in flue gas,carbon content in fly ash,water level in the drum,main steam temperature,ash deposition on economizer and pollutant emission,and others.The results show that,for parameters showing nonlinear characteristics like the carbon content in fly ash,the soft measurement model established by the kernel principle component analysis（KPCA）method has well effect.For the parameters with high linear related degree between each other,the partial least square regression（PLSR）method is more efficient to build up the their soft measurement models.The model established by the ANN method has larger output errors when the actual sample space is out of the training sample space area.So during practical engineering process,the parameters of the model built by the ANN method should be corrected periodically.The SVM method still has no mature guidance to establish parameters soft measurement model,using the experiential data will affect the model accuracy significantly.Compared with the conventional SVM method,the least squares support vector machine（LS-SVM）method has shorter training time and more accurate results,which is more suitable for online modeling.For fuzzy theory method,the accurate mathematic model of the measured object is not necessary,but the fuzzy system has no learning function itself.If we combining the fuzzy theory with other artificial intelligence method like the ANN method,the soft measurement performance will be enhanced.Therefore,the application of soft measurement technology makes possible the effective measurement of thermal parameters.