Research On Iterative Learning State Estimation Methods For Nonlinear Batch Process

Batch processes are characterized by the same production procedure for different batch runs,and they are widely used in the production of food,biopharmaceutical and chemical industry.There are some key parameters(also known as state variables)which cannot be measured online or can be measured online but with high cost in the batch process.These key state variables are particularly important for the stable and optimal operation of the production process and product quality.Therefore,the state estimation of the batch process is the focus of the industry and academia.Kalman filter(KF),as a classical state estimation algorithm,estimates the real-time measurement data and the system state along the time direction.Considering the characteristic of multi-batch runs,iterative learning Kalman filter(ILKF)not only estimates the state error between two adjacent batches along the time direction,but also updates the current state estimation along the batch direction.As a result,the two-dimensional characteristics of time and batch direction is taken into account.However,this method is only suitable for linear systems.For the nonlinear,multi-stage and multi-batch characteristics of batch process,the state estimation method of nonlinear batch processes based on iterative learning is studied by combining with the existing state estimation methods.The specific research results of this thesis are as follows:(1)For nonlinear batch processes,an iterative learning quasilinear Kalman filter(ILQKF)method is proposed to capture the nonlinear feature and the iterative feature.Based on the nominal model of batch process,ILQKF takes the error between the actual state and the nominal state as a new state and establishes a linear model for the error transition.Then,the nominal trajectory is used to estimate the state of the nonlinear batch process,and the error of state estimation gradually convergence with the increasing number of batch runs.(2)Based on the state augmentation method,to improve the estimation accuracy,an iterative learning augmented quasilinear Kalman filter(ILAQKF)method is proposed to incorporate online delay-free measurement with low accuracy and offline delayed measurement with high accuracy into the state estimation.Firstly,the nonlinear system is quasilinear by a nominal model,and the extended model is established through the augmented matrix method.Then,the ILAQKF is designed based on ILQKF.The simulation results show that the proposed method is better than the case without delayed measurement.(3)For a nonlinear batch process without a state-space model,the iterative learning state estimation method is proposed based on a linear multiple model.Firstly,the linear parameter varying(LPV)model is used to approximate the complex nonlinear process with multiple simple local linear models.Then a multi-LPV(MLPV)modeling method is used to describe the multi-batch process.The MLPV model is extended to state space(SS)model.Considering the repeated interference among batches,an error system is constructed.Finally,an improved ILKF scheme is designed to estimate the error and indirectly obtain the estimation of states,which achieves the state estimation of the batch process with unknown model.