Multimode Process Fault Detection Approach Based on IGSA-KPCA Neighborhood Modeling

In order to improve multimode process fault detection low accuracy,an ensemble method called improved gravitational search algorithm-kernel principal component analysis( IGSA-KPCA) neighborhood modeling is proposed. Firstly,the related data is found in reference data sets by using just in time learning( JITL) approach,then the related data is set and current data are used as inputs of the KPCA model. KPCA model parameters have great influence on fault detection performance and improved GSA is put forw ard to optimize the KPCA model parameters,which improves fault detection performance. Finally,the proposed method is applied to penicillin multimode process and the simulation results show that IGSA-KPCA neighborhood modeling method is better than traditional method for multimode process fault detection with fast and high accuracy.