基于模糊c均值聚类的多模型软测量建模

Study on soft Sensing Modeling via FCM based Multiple Models

根据几个模型相加可提高模型的预测精度及鲁棒性的思想，提出了一种非线性软测量建模的新方法。即先用模糊ｃ 均值聚类算法将训练集分成具有不同聚类中心的子集，每一子集用ＲＢＦ网络或部分最小二乘法进行训练得出子模型，再用模糊聚类后产生的隶属度将各子模型的输出加权求和得到最后结果。此算法通过一个复杂非线性函数的仿真建模和一个分馏塔柴油倾点软测量建模的工业实例研究，结果表明比其它算法具有更好的泛化结果和预报精度，具有良好的在线应用潜力。

Inspired by the idea of combining models to improve prediction accuracy and robustness,a new method for nonlinear soft sensing modeling of chemical processes is proposed.Fuzzy c means clustering (FCM) algorithm is used for separating a whole training data set into several clusters with different centers,each subset is trained by radial base function networks (RBFN) or partial least square algorithm (PLS). The degrees of membership is used for combining several models to obtain the finial result. The proposed method has been evaluated by a nonlinear function example and applied to a practical case of modeling product quality of hydrocracking fractionator. The obtained results demonstrate the promise of this approach for improving nonlinear soft sensing modeling.