The Study On The Hybrid Soft Sensor Modeling Of Bisphenol A Production

Process modeling is the foundation of optimal control for industrial processes. Hybrid modeling combine the advantages of both empirical modeling and mechanistic modeling, which not only reflects the main laws of the actual system and uses priori information, but also considers the impact of unknown disturbance and uncertainty. So the hybrid model has higher accuracy than empirical model and higher reliability than mechanistic model. Taking the Bisphenol A (BPA) production as the object, this paper studies the theory and application of hybrid modeling.Based on the study of BPA reaction mechanistic, this paper builds the hybrid model combining the mechanistic model and support vector machine (SVM). The reaction mechanistic is used as the model framework to describe the reaction trend. Then support vector machine is applied to estimate the unknown process parameters according to production situation. The online estimation of BPA content at the reactor outlet is realized. Simulation results confirm the accuracy of the model.In the Bisphenol A production the cation exchange resin is used as the catalyst. The catalytic activity has an important impact on the BPA production. On the basis of studying the mechanistic of catalyst deactivation, this paper builds a semi-empirical formula to describe the catalytic activity trend, and then uses support vector machine to estimate the fluctuation of catalyst activity according to the production situation. The online estimation of catalyst activity is realized. Simulation results confirm the accuracy of the model.Data preprocessing is important for soft sensing, which affects the performance of soft sensors directly. To avoid the negative impact of noised data, this paper proposes a method of data preprocessing combining the Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) algorithms to eliminate the noises in the data to improve the performance of the soft-sensing model and its estimation accuracy. Simulation results confirm that the data preprocessing method is feasible and effective.The research findings of this paper have been applied to actual production processes and passed the identification of the Ministry of Education with satisfactory results.