Multi-model Switching Method Based on Sphere-Based SVM Classifier Selector and Its Application to Hydrogen Purity Multi-model Soft Sensor Modeling in Continuous Catalytic Reforming

The process of continuous catalytic reforming is complex and changeable. Usually, a single model soft sensor is hardly to grantee the accuracy of the prediction result, so it is necessary to adopt the multi-model strategy to improve the model performance. The process of sub model combination of the multi-model soft senor could be considered as a multi-class classification issue. The main idea of the proposed method in this paper aims to solve this issue with Support Vector Machine (SVM). The proposed approach is to build a sphere structure to cover the same-class samples as much as possible, and these sphere-based structure can be considered as a selector of those SVM classifiers. Experimental results show that the proposed method is suitable for particular use in SVM multi-class classification, and the switched-based multi-model soft sensor for hydrogen purity in continuous catalytic reforming based on the proposed method has a higher prediction accuracy.