Soft Sensing of Distillation Systems Based on Data Reconstruction and Optimized LSTM Model

The data driven soft sensing method has remarkable effects on predicting the variables of complex systems such as the distillation process. However, it has high requirements for the accuracy of soft sensing. In this work, a soft sensing algorithm based on the combination of CEEMDAN and the optimized LSTM model is proposed. Firstly, the HSIC algorithm is used to select auxiliary variables. Then, CEEMDAN is utilized to perform modal decomposition on the noisy data, so as to obtain measurement data in different frequency bands and complete data reconstruction. Finally, the optimized WOA-LSTM model is constructed. Through the verification of practical simulation comparison experiments, it is verified by different quantitative indicators that the proposed soft sensing method has high prediction accuracy and stability.