Systematic estimation of state noise statistics for extended Kalman filters

摘要: The successful application of model-based control depends on the information about states dynamic system. State-estimation methods, like extended Kalman filters (EKF), are useful for obtaining reliable estimates from a limited number measurements. They also can handle model uncertainties and effect unmeasured disturbances. main issue in applying EKF remains that one needs to specify confidence terms process noise covariance matrix. effectively systematically calculate matrix an EKF. Two systematic approaches used this calculation. first is based Taylor series expansion nonlinear equations around nominal parameter values, while second accounts dependence system fitted parameters by Monte Carlo simulations easily be performed on-line. value obtained not diagonal form current state Thus a-priori regarding uncertainty utilized need extensive tuning eliminated. these techniques example processes discussed (Transesterification process; Methyl methacrylate polymerization; Semibatch using tendency models). accuracy methodology compared very favorably with traditional methods trial-and-error