State Estimation Methods Based On Particle Filter In Batch Process With Delayed Measurements

Batch processes are widely used in various industries and play an important role in the development of national economy.Limited by sensor detection technology,the measurements of online analyzers is often infeasible.Since the batch process is characterized by strong nonlinearity,non-Gaussianity and time-varying,it is difficult to monitor and optimize the production process.Based on the state space model,the state estimation method can accurately estimate the key variables in the form of filtering according to the statistical law of the measurements.In the actual industrial processes,there exists two kinds of measurements: online measurements are delay-free with low accuracy;offline measurements have time delay with high accuracy.On the other hand,a typical batch process has the characteristics of periodic batch production and contains many batches.This dissertation mainly studies the state estimation in batch processes with the consideration of the delayed measurements and multiple-batch characteristics.This dissertation includes the following parts:(1)In batch processes,the key variables are usually obtained online with low accuracy or offline with large time delay,and a state estimation algorithm is proposed to estimate the key variables by incorporating delayed measurements with the real-time measurements.Due to the different sampling intervals of these two kinds of measurements,two cases are analyzed,including the case of only real-time measurements available and the case of both real-time and delayed measurements available.The particle filter algorithm is introduced for the state estimation,and it is further extended by the Bayesian method for the information fusion of these two kinds of measurements.Finally,the proposed method is applied in a numerical example and the beer fermentation process and obtains good results.(2)Considering the characteristic of multiple batches and the information of delayed measurements,a two dimensional state space model is developed and a two dimensional state estimation algorithm is proposed.In the proposed algorithm,information of previous batches and delayed measurements is fused by using Bayesian method and the forward-backward smoothing algorithm.The estimation results get better with the increase of the batch dimension and incorporation of delayed measurements.The applications in a numerical example and a beer fermentation show the effectiveness of the proposed method.(3)In order to solve the problem that the two dimensional state space model is difficult to be established accurately,this chapter proposes a new algorithm based on iterative learning according to the repetitive characteristic of batch processes.Based on the estimated state from previous batches in the same sample time and the measurement model,the expected measurements are estimated and the tracking error is obtained from the difference between the actual measurements and expected measurements.For the delayed measurements,the Bayesian method and the forward-backward smoothing are used to fuse the information.Finally,the effectiveness and practicability of the proposed method are verified by a numerical simulation and the beer fermentation process.