Data Assimilation for Combined Parameter and State Estimation in Stochastic Continuous-Discrete Nonlinear Systems

Abstract

Data assimilation (DA) provides a general framework for estimation in dynamical systems based on the concepts of Bayesian inference. This constitutes a common basis for the different linear and nonlinear filtering and smoothing techniques which gives a better understanding of the characteristics and limitations of each approach. In this study, four nonlinear filters for systems governed by stochastic continuous dynamics with discrete measurements are inferred as different approximate solutions to the Fokker-Planck equation in the prediction step of the Bayesian update. The characteristics and drawbacks of each filtering approach are discussed in light of the introduced approximations to the original Bayesian problem in each case. The introduced continuous-discrete (CD) filters are then implemented to solve the combined estimation problem of the state and parameter variables in a nonlinear system described by a stochastic differential equation (SDE), namely, the continuous stirred-tank reactor (CSTR). The performance measures for each filter, such as estimation errors, computational cost, and the ability to quantify the correct uncertainty, are compared and investigated with attribution to the assumed approximations introduced in the derivation of each class of filters.