Multi-model dynamic fusion soft-sensing modeling and its application

In this paper, a multi-model dynamic fusion soft sensor modeling method based on Gauss-Markov estimation is proposed. Firstly, the fuzzy c-means algorithm is used to cluster the input samples of the model. The radial basis function and least square support vector machine are used to establish multiple sub-models for each clustering. The multi-model outputs are predicted by dynamic fusing the values of sub-models based on the Gauss-Markov estimation. The proposed method is applied to predict alumina powder flow in the process of alumina conveyor. The results indicate that the proposed method has higher predictive accuracy and better generalization capability in comparison with the other soft sensor methods.