Research On Intelligence Modeling, Control And Optimization For Poly (Ethylene Terephalate) Process

Poly (Ethylene-Terephthalate) (PET) is a kind of polymers which is widely used in industrial process and daily life. Production process of PET is highly non-linear, slow time-varying and with distributed parameters. With severe competition in PET market, profit improvements of running optimization in industrial processes attract much attention. Artificial intelligence methods such as support vector machines, Gaussian processes and evolutionary algorithms have been applied in the domain of chemical process engineering. These methods solved some control and optimization problems of complex chemical system. However, as a statistic method, optimizations of Gaussian process parameters have complex computation; estimation of distribution algorithm, a hot spot of evolutionary algorithm, is faced with the selection of probability model and easiness to trap into local optimals. The wide application of intelligence methods still needs further research and development.This paper focuses on industrial problems of large-scale PET production process and emphasizes on the theoretical and technical research on applying artificial intelligence method to modelling, control and optimization. Running optimization software of PET process has been developed. All of these provide a new way for PET process operation optimization. This paper covers the following parts:For the problem that probability model of population is hard to be determined during evolution of estimation of distribution algorithm, single and multi-objective kernel density estimation of distribution algorithms were proposed. The selection criterial of kernel width has also been discussed. Numerical simulation results validated the effectiveness of the algorithms. The algorithms were applied to optimize kinetic parameters of PET process with plant operating data. The industrial kinetic model has been obtained and its efficiency were validated via comparing with industrial data. Based on mechanism, whole process model of PET was developed and certificated with industrial data.In this paper intelligent modeling method of ethylene glycol which is raw material of PET was discussed. Priori knowledge was used into support vector machine and support vector model with input variables monotone was proposed. This method was used for modeling of catalyst deactivation and intelligence modeling of oxidation reactor was implemented. Kinetic parameters were optimized with plant data and model of hydration reactor was developed. Whole process of ethylene glycol were simulated and validated with industrial data.In order to find cumulative effection of process fluctuations, disturbance and mutative conditions, dynamic model of esterification and final polycondensation process were developed. Step responses of end carboxyl concentration and ethylene flow rate to feed molar ratio, temperature, pressure were discussed and dynamic characteristics of esterification process were obtained. Dynamic model of polycondensation process was established using series reactors and step responses of intrinsic viscosity to vacuum, temperature were analyzed.This paper discussed gaussian process for soft sensor modeling. Clustering based sparse gaussian process was proposed in order to reduce the complex computation of gaussian process. It was used in soft sensor modeling of end carboxyle concentration in esterification process and the model gave the prediction and mean variance. Pseudo-input sparse gaussian process was used in this paper and online learning method was added. It is used in the modeling of b value of PET. Also gaussian process was also used in prediction control. Prediction control based on gaussian process was used for the control of instrinsic viscosity in final polycondensation process.For the optimization of PET process, hybrid optimization algorithms based on estimation of distribution algorithm, Cauchy distribution and particle swarm optimization were proposed. Based on the whole process model of PET, energy consumption was optimized using hybrid optimization algorithm. The optimal operation were obtained and used in industrial process for energy-saving.Based on service-oriented architecture and multi-agent, modeling, control and optimization were integrated into running optimization framework. Basic services in industry information system were defined and interactions between services were discussed. The platform of running optimization was developed and integration software of PET modeling, control and optimization were established.