A Soft Sensor Based on Multiple Neural Networks Combined with Two Information Fusion Methods

Soft sensors are especially required in lots of advanced process control applications. The ANN based soft sensor are widely studied recently. But the ANN is an uncertain method in nature. In view of the complexity of the industrial processes, the robustness is an important criterion to evaluate a model. The generalization capability is another factor to affect the applicability of a model. Aiming at improving the robustness and generalization capability of a system, a two-level architecture MNN model is proposed for soft sensor modeling. In our model, multiple networks are combined with the Bayesian and fuzzy C-means (FCM) clustering combination methods at different levels. Two experiments are conducted to validate the effectiveness of our model. The results reveal that the proposed model exceeds other three models indeed