基于MMI-PCA-KLPP二次降维和模糊树模型的NO_X浓度软测量方法

针对火电厂气体分析仪测量存在滞后和NO_X软测量不准确的问题,提出了一种基于改进互信息-主成分分析-核局部保持投影(MMI-PCA-KLPP)二次降维和模糊树模型(FT)的NO_X浓度软测量方法。首先对样本进行剔除离群点等预处理,再通过查阅文献大体确定模型的输入变量,并采用MMI方法对输入变量进行降维处理;针对MMI降维后依然有较多输入变量的问题,综合考虑了样本全局结构特性和局部结构特性,用PCA-KLPP方法对MMI降维后的变量二次降维;最后针对二次降维后的数据,基于模糊树算法建立了NO_X软测量模型。实验结果表明,模糊树模型精度高且泛化能力强,结合MMI-PCA-KLPP二次降维处理后,大大缩短了模型的训练时间且精度未出现较大程度的降低。

Soft Measurement Method of NOx Concentration Based on MMI-PCA-KLPP and Fuzzy Tree Model

Aiming at the lag and inaccurate NO_X soft measurement in gas analyzers in thermal power plants, this paper proposed a NO_X concentration soft measurement method based on improved mutual information, principal component analysis and core local hold projection(MMI-PCA-KLPP) secondary dimension reduction and fuzzy tree model(FT). Firstly, it preprocessed the samples such as outliers, determined the input variables of the model by consulting the literature and reduced the dimension of the input variables by MMI method. But there were still many input variables after dimension reduction. Thereof, it reduced the dimension of the MMI after dimension reduction by PCA-KLPP method considering the global structural characteristics and local structural characteristics of the sample. Finally, it constructed NO_X soft-measurement model by fuzzy tree algorithm based on the data after the second dimension reduction. The experimental results show that FT model possesses high precision and strong generalization. Combined with MMI-PCA-KLPP secondary dimension reduction, the training time of the model is greatly shortened and the accuracy is not obviously reduced.