Research On Parameter Identification And Soft Sensor Method For Propylene Oxidation Process

Propylene oxidation process is an key part of acrylic acid production process and has been adopted widely in the world.Propylene can be effectively translated into acrylic acied by acrylic acid production process,which can increase acrylic acied yield.Protection of catalyst,explosion protection,the percent conversions of propylene and acrolein directly affect the lifetime of catalyst and various reactors,yield of acrylic acid.Therefore,it is significance to ensure the safe and stable operation of the propylene oxidation process.The background of this dissertation is the actual production process of acrylic acid.To deal with the unknown parameters of catalyst protection region,multi models and multi unknown parameters of propylene explosion region,real-time measurement of the propylene and acrolein conversion,the parameter identification and soft sensor methods are developed.The main contributions of this dissertation are fivefold:(1)To deal with parameter identification problem of catalyst protected region in the process of propylene oxidation,this dissertation proposes a novel parameter identification method based on an adaptive simulated annealing particle swarm optimization(ASAPSO)algorithm.Through linear decreaseing weights and synchronous change learning factors,search ability and information exchange capacity are improved.Compared with other methods,the proposed method has some adventages,such as high accuracy,fast convergence and good stability.Finally,the corresponding simulation verification is carried out.(2)Considering the parameter identification problem of multi models,multi unknown parameters in the propylene explosion region,a new parameter identification method for propylene explosion region based on self-adaptive chaotic local search particle swarm optimization(SACLSPSO)algorithm is proposed.The particle swarm optimization(PSO)is improved by chaotic local search and nonlinear dynamics inertia weight coefficient.Through weihted average method,multiple propylene explosion functions are transformed into a total objective function.Finally,the proposed method is applied to identify the parameters of propylene explosion region.In order to verify the reliability of the identification parameters,the corresponding simulation verification was carried out,and the propylene explosion areas of the reactor and mixer are obtained.(3)The experience model of propylene explosion is linear.Nevertheless,there is error between the actual limit of propylene explosion and experience model.To deal with nonlinear relationship between the volume concentration of feed mixer inlet gas and propylene explosion limit,a new soft sensor model of propylene explosion limit based on kernel particle least squares(KPLS)is proposed.Then,propylene explosion region is obtained.(4)As the acrolein is hard to be real-time measured,a novel soft sensor model of acrolein conversion based on hidden markov model(HMM)with particle component analysis(PCA)and fireworks algorithm(FWA)is proposed.The HMM is improved by PCA and FWA.The HMM is trained and predicted by Baum-Welch algorithm and Viterbi algorithm.Finally,by using weihted average method,the result of soft sensor is obtained.The simulation results show that the proposed soft measurement method can predict the acrolein conversion rate.(5)Considering the problem of off-line measurement for propylene conversion,a new soft sensor modeling method of propylene conversion based on T-S fuzzy neural network optimized by independent component analysis(ICA)and mutual information(MI)is proposed.The method can achieve denoising,key variable selection,and reduce the influence of weak correlation,so as to better predict the propylene conversion rate during propylene oxidation.Simulation results show that the proposed soft sensor modeling method has better prediction accuracy and smaller error.