Soft Sensor Modelling Based on Mutual Information Variable Selection and Partial Least Squares

To remove redundant variables and resolve the high correlation problem of soft sensor modelling, this paper proposed an efficient mutual information(MI) based partial least squares(PLS) method. First, we use MI criterion to sort the variables in a descending order according to their importance. Then the linearity between process variables and quality variables is tested through F test. If there is a strong linearity between them, then use PLS method to build regression models between different number of process variables and quality variables and find the final selected variables according to the minimum root mean square error(RMSE). The proposed method can enhance the robustness and interpretability of the soft sensor model, reduce the amount of calculation, and improve the precision of prediction. The efficiency of the proposed method is demonstrated through two case studies by comparing with other four variable selection methods.