Abstract: According to the characteristics of the boiler combustion process with nonlinear, multiple operating regions and multivariable coupling, a multiple-model modeling method based on fuzzy C means clustering and least squares support vector machines and weighted connections (FCM-LSSVM-WC) is proposed. The influence of inputs on outputs is employed to evaluate the difference between the samples. A BP neural network with "limited" processing is used to calculate the MIV. The membership weights are the MIVs which are used to realize classification and connect the multiple model. The proposed method is verified by taking the circulating fluidized bed boiler on a thermal power plant. Industrial applications show that compared with PLS, LSSVM, FCM-LSSVM, AP-LSSVM, the modeling method can ensure the generalization accuracy requirements, simultaneously possess better tracking ability in predicting NOx emissions of the boiler combustion process.

Key words: Boiler combustion, average impact value, fuzzy C means clustering, multiple model, NOx emissions

摘要： 针对热电锅炉燃烧过程存在非线性、多工况、多耦合等特点，提出了一种模糊C均值聚类、最小二乘支持向量机和加权连接（FCM-LSSVM-WC）相结合的多模型建模方法。该方法根据输入变量对输出变量的影响程度来评价样本之间的差异性，利用限幅处理的BP神经网络计算平均影响值（MIV），并以此作为FCM聚类过程与多模型连接的权系数，从而保证了建模过程的连贯性与统一性。以循环流化床（CFB）锅炉为例开展燃烧过程NOx排放量的建模仿真研究，结果表明：与偏最小二乘算法（PLS）、LSSVM算法、FCM-LSSVM算法、仿射传播-最小二乘支持向量机（AP-LSSVM）算法相比，该建模方法能够在保证泛化精度的同时，具有更好的跟踪预测能力。

关键词: 锅炉燃烧, 平均影响值, 模糊C均值聚类, 多模型, NOx排放