Using a single model to describe a complex nonlinear object

it usually suffers from low accuracy and poor generalization.A multiple model soft sensor approach is presented based on local reconstruction and fusion manifold clustering.In order to restraining the impacts of outliers to clustering results

the data set is split into several small disjoint sub-clusters.By reconstructing linear manifold level based on every sub-cluster respectively

the sub-clusters which are closer and in the same manifold level are merged.Meanwhile

support vector machine is used to construct regression model in terms of each sub-class and a soft-sensor composed model based on the multiple sub-models is obtained finally.The proposed algorithm is used in a soft sensor modeling for the Bisphenol-A productive process

and the result of simulation shows the effectiveness of the algorithm.