针对实际工程中噪声难以避免和预测的问题,提出了鲁棒性较强的加权模糊树(W-FT)算法,采用基于局部异常因子(LOF)的加权最小二乘法代替最小二乘法学习模糊规则的后件参数,通过2个典型的非线性例子验证了该算法的有效性.应用W-FT算法建立了电站锅炉NOx排放特性模型,并与其他建模方法所建模型进行了对比.结果表明:所提出的W-FT算法能够有效地辨识噪声和异常值,具有较强的鲁棒性,所建立的模型预测精度较高,泛化能力较强.%Aiming at the problem that the noise is unavoidable and difficult to predict in actual engineering projects,a weighted fuzzy tree (W-FT) algorithm was proposed by utilizing the weighted least squares method based on local outlier factor (LoF) to replace the ordinary least squares method to learn the consequent parameters of the fuzzy rules,which were subsequently validated by two typical nonlinear examples.Meanwhile,a soft sensor model was established based on W-FT for NOx emission of boilers,and was then compared with other modeling methods.Results show that the proposed W-FT algorithm can effectively recognize the noise and outliers,and the models built on the basis of W-FT have higher prediction accuracy and stronger generalization capacity.
展开▼
