Research On Soft-sensor Modeling Method Of SMB Chromatographic Separation Process

Aiming at the problems of low accuracy and poor real-time performance of the traditional manual detection method when measuring the simulated moving bed(SMB)chromatography separation process,soft-sensor modeling method is adopted to realize the prediction of the key indexes of the simulated moving bed chromatography separation process.Therefore,the soft-sensing methods based on fuzzy neural network(FNN)were proposed for realizing the real-time control in the moving bed chromatography separation process.The main work of this thesis can be summarized as follows.Firstly,based on the technique of simulated moving bed chromatographic separation process and the application characteristics of the adaptive neural fuzzy inference system(ANFIS),a soft-sensing model was established.By adopting the grid partitioning,subtraction clustering and fuzzy C-means(FCM)clustering algorithm,the input data space of adaptive neural fuzzy inference system is divided.Then the gradient algorithm,Kalman algorithm,Kaczmarz algorithm and Pseudoinv algorithm were used to optimize the parameters of ANFIS.A hybrid algorithm combining data space partition method and post-parameter parameter optimization algorithm is used to optimize the ANIFS.The proposed soft-sensing model based on ANFIS was established to predict the key index of the chromatographic separation process in a simulated moving bed.Secondly,the particle swarm optimization(PSO)algorithm is used to optimize the parameters of the adaptive fuzzy inference system.The different inertial weight adjustment methods on PSO algorithm are studied.The performances of the optimized adaptive fuzzy inference system are compared.Based on the PSO algorithm with random inertia weights,the inertia weights of the function on the number of iterations,and the inertia weights of the feedback information function,the ANFIS soft-sensing model is optimized by adopting a hybrid learning algorithm based on a combination of an improved adaptive population evolution particle swarm optimization(NAPEPSO)and the least-means squares(LMS)method to optimize the model parameters.The validity of the proposed soft-sensor model is verified by simulation experiments.Finally,aiming at the problem of too many rules in the first two kind of soft-sensor models,different parameter adjustment methods of the dynamic fuzzy neural network(D-FNN)are studied.By using Kalman filter(KF)algorithm,linear least squares(LLS)method and extended Kalman filter(EKF)method,a dynamic fuzzy neural network soft-sensor model is established.The sliding window algorithm is used to realize the adaptive revision on the D-FNN soft-sensing model for the moving bed chromatographic separation process.In conclusion,the simulation results show that three fuzzy neural network soft-sensing models have achieved good prediction results.It can improve theaccuracy and robustness of economic and technical index prediction in the simulated moving bed chromatography separation process,and can meet the real-time control requirements of the simulated moving bed chromatography separation process.