Parameter Estimation Based On Extended Kalman Filter

Parameter estimation is an important research filed in engineering practice.Generally,a mathematical model of the practical system is usually established for theoretical analysis.Parameter estimation is the core of modeling,and it is also the primary premise of the system control and optimization.There are often interferences in practical process,which increases the difficulty of parameter estimation.The accuracy of parameter estimation determines our cognition of the practical system,and also affects the design of system control and optimization.It is worth to focus on the research of parameter estimation.As for the single rate soft sensor system,the parameter estimation of time-invariant and time-variant system are studied separately.When parameters are time-invariant,the linear model is obtained after some transformations so that the Kalman filter(KF)can be used for parameter estimation.The simulations based on extended Kalman filter(EKF),KF and stochastic Newton recursive algorithm(SNR)are shown for comparison,and the accuracy of KF is the best.When parameters are time-variant,the KF parameter estimation with a covariance modification(CM-KF)is proposed to track the time-variant parameters,and the simulations show the good tracking performance.As for the dual-rate soft sensor system,the parameter estimation of time-invariant and time-variant system are studied separately.When parameters are time-invariant,the EKF algorithm with an adjustment factor ? is proposed for better accuracy and the convergence analysis is presented through associated differential equation theory.The influences of the adjusting factor ? to convergence rate and estimation accuracy are analyzed,and EKF parameter estimation algorithm with a slower convergence rate(S-EKF)is presented as well.As for time-variant parameters,the S-EKF algorithm with a covariance modification(CMS-EKF)is proposed for better tracking performance,and the simulations have shown the effectiveness.As for a class of canonical dynamic system,the parameters are augmented to the state variables and a nonlinear system is obtained.The parameters to be estimated and the original state variables are both included in state variables of the nonlinear system.EKF is used for parameter estimation of the nonlinear system and the convergence is analyzed by constructing associated differential equation.Finally,the simulation demonstrates the good performance.