Adaptive soft sensor modeling method based on multi-model dynamic fusion and its industrial application

In soft sensor modeling, a single model can not accurately describe the complex nonlinear object, and multi-models use static model in many cases, thus the dynamic changes in actual operation of the system are almost not being considered. To overcome this problem, an adaptive soft sensor modeling method based on multi-model dynamic fusion is proposed. In this method, the training samples are first clustered by the affinity propagation algorithm, and then by training each clustering with the Gaussian process regression algorithm, a sub-model is built for each clustering. At last, the predicted values of the sub-models are dynamically fused by adaptive Gauss-Markov estimation. The proposed method is applied to predict the p-xylene (PX) purity in the adsorption separation process. Results indicate that the proposed method actually increases the model´s adaptive abilities to various operation conditions and improves its generalization capability.