Study Of Hybrid Modeling Method And Its Application In Chemical And Engineering Processes

The soft-sensor of process modeling plays a vital role in chemical industry,it is an effective means to improve industrial process control,optimization and measurement of complex chemical properties.Because the actual process is characterized by high complexity,nonlinearity,severe coupling and time-varying,the mathematical model must not only be able to accurately fit the steady-state characteristics of the process,but also be usd to reply a wide range of process dynamics.Hybrid modeling shows the mechanism and characteristics of complex processes,and the influence of unknown disturbances or uncertain factors on the actual system.It can improve the prediction accuracy and reliability of the model effectively.This paper introduces the research of hybrid modeling methods,and the model studied is used to achieve the solubility prediction of CO₂ in ionic liquids and the soft-sensor of melt index in propylene polymerization process.The main research contents are as follows:?1?After collecting a large number of related literatures at home and abroad,this paper introduces the modeling problems of complex chemical processes and the research background of the system industry.It summarizes the basic principles,content and research status of three process modeling methods:mechanism modeling,data-driven modeling and hybrid modeling.According to the structural design method of hybrid model,the basic idea,development history,design principle,application field and application status of hybrid modeling technology are introduced,and commented on their strengths and weaknesses.?2?A semi-empirical and semi-mechanical series structure hybrid modeling method was proposed to predict the solubility of CO₂ in ionic liquids.The Krichevsky-Kasarnovsky mechanism model reflecting thermodynamic properties was established by mechanism analysis to construct the overall structure describing the characteristics of vapor-liquid equilibrium process.On this basis,the Henry's constant empirical Valentiner and infinitely diluted partial molar volume empirical polynomial are used to correct the missing part of the model mechanism,and the output of the empirical model is constrained by the prior conditions of the experimental measurement process to satisfy the phase equilibrium with the aim to satisfy the intrinsic mechanism knowledge of the phase equilibrium process,and to construct a series structure type hybrid model;Then,according to the experimental data collected in the literature,the Levenberg-Marquardt optimization algorithm is used to learn and modify the parameters of the constructed hybrid model to improve the application range and prediction accuracy of the model.The model was used to predict the solubility of CO₂ in ionic liquids.Through the model performance test,the prediction results were consistent with the corresponding experimental data,reflecting the good prediction performance of the model.?3?A parallel structure of hybrid modeling method based on mechanism model and dynamic error compensation model was proposed to realize online soft-sensor of melt index in polypropylene production process.According to the polymerization mechanism analysis,the direct mechanism model of the melt index of propylene polymerization process was established.Based on this,the Elman neural network is used to construct N sub-models,and the fuzzy C-means clustering algorithm is used to obtain the sub-models with better performance to construct the data-driven model to realize the error compensation of the predicted value of the intuitive mechanism model.It is mainly considered that the polymerization process is prone to the characteristics of working point drift,multimodal or frequent working condition switching.At the same time,the correction bias of the error compensation model is judged,and the correction mechanism model parameters are updated in time to improve the prediction accuracy of the hybrid model.This modeling method is applied to the soft-sensor prediction of polypropylene melt index,which can accurately predict the change trend of melt index in polymerization process,and compared with single mechanism model or single data-driven model,the hybrid model has better predict performance.The hybrid modeling method developed in this paper is applied to the soft-resistance study of ionic liquid absorption CO₂ and propylene polymerization process.According to its different process characteristics,a hybrid model structure with two different process information sources is designed and studied,and better application results are obtained.The current research work has a certain application value and reference significance for the in-depth study of chemical and engineering soft-sensor modeling methods,which helps to understand the nature,characteristics and dynamic trends of system processes,It can also provide an effective method basis for process simulation optimization and process control research.