Research On Intelligence Modeling, Control And Optimization For P-Xylene Oxidation Process

Terephthalic acid (TPA) is one of the main raw materials used to produce polyester, including polyester fibre, polyethylene terephthalate (PET) bottle resin and polyester film. In industry, most of TPA is produced by the liquid-phase catalytic oxidation of p-xylene (PX). The reaction mechanism is extremely complicated, involving many coupling effects. This paper focuses on modeling, control and optimization of industrial p-xylene oxidation reaction process on the basis of research of mechanism analysis of this process. Based on the mechanism models, process simulation, control and optimization are carried out by using intelligence methods. All of these provide a new way for large and complex industrial p-xylene oxidation reaction process optimization and control. This paper covers the following parts:(1) Research on industrial intelligence modeling methodThe steady-state model are widely used in the model identification, optimization, fault detection and control of the industry process. The measurements usually contain gross errors, and gross errors existing in the industrial process will greatly affect the performance of the steady-state detection, giving the wrong or unsatisfactory results. From both economic and control points of view, data processing such as steady-state identification and gross-error detection becomes an important task and is often used for online process monitoring, product property prediction, etc. Due to the disadvantages of the traditional steady-state identification (SSI) methods, the adaptive polynomial filtering (APF) method is used for SSI in this paper. Furthermore, the APF steady-state identification with the new3d formula method is modified for gross errors detection by using the quartile method based on first order differential in this paper.Prior knowledge widely exists in industrial process. The weighted least square support vector machine (WLS-SVM) method has the ability in quick computing and the support vector sparsity. The presence of gross errors can corrupt a model’s performance, giving undesirable results. A novel weighted least square support vector machine regression (WLS-SVM) is proposed, which combines gross error detection and adaptive weight value for the training samples. First, the3δ principle is applied to detect the gross error. Second, the initial weight is obtained according to the fitting error of each sample. Then, an adaptive immune algorithm (AIGA) is applied to obtain the optimal parameters of the WLS-SVM. Furthermore, the AIGA-WLS-SVM method is applied to estimate the rate constants of an industrial p-xylene oxidation model, and the satisfactory result is obtained.(2) Intelligence modeling for p-xylene oxidation processSince the p-xylene oxidation process is very complicated, the mechanism of the process is deeply investigated. Detailed analysis of the influence of catalyst concentration, temperature, water concentration and solvent ratio are given based on the process. A novel kinetic model based on free radical mechanism is used to simulate the oxidation of p-xylene (PX) in continuous stirred-tank reactor (CSTR) under industrial operating conditions. Furthermore, correction coefficients are introduced to accurately evaluate the kinetic parameters based on Adaptive Immune Genetic Algorithm (AIGA) due to the significant difference between the nature of PX oxidation conducted in the laboratory SBR and in the industrial CSTR. In addition, the PX oxidation reaction involves many side reactions. By adding the side reactions, the extended kinetic model based on the free radical reaction mechanism is established. An adaptive differential evolution (ADE) is proposed to correct the rate constants and the industrial p-xylene model is obtained.(3) Intelligence optimization forp-xylene oxidationThere have been many attempts to optimize operating conditions of the reaction process using intelligent methods. As a new branch of evolutionary algorithm (EA), the immune techniques are widely concerned by researchers in many applications. In immune genetic algorithm (IGA), the values of the mutation factor are usually fixed or change together according to a function of the individual’s current generation number. However, IGA with deterministic mutation factor suffers from the problem of premature convergence. In this paper, a modified self-adaptive immune genetic algorithm (MSIGA), in which immune concepts are applied to determine the mutation parameters, is proposed to strengthen the searching ability of the algorithm and maintain population diversity. This proposed algorithm is successfully applied to obtain the optimal operating conditions of the p-xylene oxidation process from an economical point of view.(4) Dynamic simulation and control for p-xylene oxdiation processp-Xylene oxidation process is a multivariable, strong-coupling, nonlinear and complicated system. On the basis of the steady-state model established previously, a dynamic model with PID controllers is established and the dynamic responses of the process have been studied. Due to the deficiency of the traditional single input single output (SISO) PID control, the multivariable model predictive control is implemented on the dynamic model. Simulation results of the control system show that MPC based on mechanism model as predictive model, can greatly reduce production fluctuation and realize a smoothrunning process.