基于互信息的辅助变量筛选及在火电厂NOx软测量模型中的应用

辅助变量的选取是软测量建模中重要的一步;但由于待选变量数目多、与主导变量非线性相关、信息冗余大等因素导致辅助变量的选择不够合理。在信息熵和互信息理论基础上,改进IBF和MIFS变量筛选算法,综合考虑了辅助变量和主导变量之间的最大相关性,以及辅助变量之间的最小冗余性。作为算例使用改进后的算法,筛选了某燃煤机组运行历史数据,建立了省煤器出口NOx浓度的GA-BP软测量模型。实验证明这种基于互信息的变量筛选方法可以有效提高模型的输出精度和泛化能力。

Variable Selection Method Based on Mutual Information and Its Application in Power Plant NOx Soft Sensor Modeling

The selection of auxiliary variables is an important step in building soft sensor model. However, due to the number of variables to be selected, nonlinear correlation, information redundancy and other factors lead to the selection of secondary variables is not reasonable enough. Based on the theory of information entropy and mutual information, the IBF and MIFS variable selection algorithm is improved. This algorithm takes into account the maximum correlation between the auxiliary variable and the dominant variable, and the minimum redundancy between the auxiliary variables. As an example, the improved algorithm is used to select the historical data of a coal-fired boiler, the GA-BP soft sensing model of NOx concentration of economizer outlet is established. Experimental results show that the method based on mutual information can effectively improve the accuracy and generalization ability of the model.