The State Estimation Under The Unknown Model Based On The UKF And Its Application In Acrylic Polymerization Process

In most complex nonlinear processes, many intermediate variables that need to be controlled can’t be measured directly, which directly affect the realization of process monitoring system. Nonlinear filtering techniques provided a strong foundation for complex system state estimation, and get a lot of attention and research. At present, unscented Kalman filter (UKF) algorithm has solved many practical state estimation problems for the nonlinear system. However, many uncertain factors are big challenge in the accurate and stable state estimation.This paper makes the acrylic polymerization process as the research background, study the UKF algorithm deeply. Researches are as follow:First, the typical nonlinear dynamic models such as continuous stirred-tank reactor (CSTR), pH neutralization process are analyzed, after two-step acrylonitrile polymerization craft was introduced. The summary of characteristics for these models provided the improving direction for filtering algorithm.Then, the typical nonlinear filtering algorithm was introduced. According to the analysis and study the specific steps of UKF, there were two major challenges in the filtering process:the unknown system model and the unknown statistical characteristics of noise.Due to the UKF can only solve the state estimation of nonlinear system which model was known, but real nonlinear systems are difficult to establish models for its complexity and uncertainty. The combination of the UKF and neural network (NN-UKF) solve such problems in the nonlinear process state estimation of process model is unknown and output is a linear combination of states. Through the simulations, it was verified that the proposed method had a better estimation result. The application of this algorithm in continuous stirred tank reactor process solved the problem of estimation of concentration and temperature when the continuous stirred tank reactor nonlinear model is unknown.The unknown statistical characteristics of noise and the poor robustness of UKF to noise information resulted in the deterioration or divergence of filtering precision. In this paper, an improved UKF algorithm is proposed based on robust Cauchy function method (CR-UKF). In order to reduce the weight of state which estimated noise are inaccurate, improve the accuracy of the UKF algorithm, real-time correcting the estimated noise by the combination weighting function based on the residual error between estimate and the actual value. Two simulation results showed that the CR-UKF algorithm was effective to improve the accuracy of state estimation of UKF under the inaccurate noise estimation. The application of the CR-UKF algorithm to the pH neutralization process not only increased the monitoring accuracy of pH of the reaction, but also improved the estimation accuracy of the ion concentration.Finally, it summarized the proposed algorithms in this paper, and had a further prospect for later works.