State Estimation for Continuous-Discrete-Time Nonlinear Stochastic Systems

Marcus Krogh Nielsen,Tobias K. S. Ritschel,Ib Christensen,Jess Dragheim,Jakob Kjøbsted Huusom,Krist V. Gernaey,John Bagterp Jørgensen

DOI: https://doi.org/10.48550/arXiv.2212.02139

2022-12-05

Abstract:State estimation incorporates the feedback in optimization based advanced process control systems and is very important for the performance of model predictive control. We describe the extended Kalman filter, the unscented Kalman filter, the ensemble Kalman filter, and a particle filter for continuous-discrete time nonlinear systems involving stochastic differential equations. Continuous-discrete time nonlinear systems is a natural way to model physical systems controlled by digital controllers. We implement the state estimation methods in Matlab, illustrate and evaluate their performance using simulations of the modified four-tank system. This system is non-stiff and the state estimation methods are implemented numerically using an explicit numerical integration scheme. We evaluate the accuracy of the state estimation methods in terms of the mean absolute percentage error over the simulation horizon. Each method successfully estimates the states and unmeasured disturbances of the simulated modified four-tank system. The key contribution is an overview and comparison of state estimation methods for continuous-discrete time nonlinear stochastic systems. This can guide efficient implementations.

Optimization and Control,Systems and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to perform state estimation in continuous - discrete - time nonlinear stochastic systems. Specifically, the paper focuses on how to effectively predict and reconstruct the state of these complex systems, especially in physical systems controlled by digital controllers. Such problems are crucial for the performance of advanced process control systems (such as model - predictive control).

To achieve this goal, the paper introduces four state - estimation methods:
1. **Extended Kalman Filter (EKF)**: Apply the equations of the original Kalman filter by locally linearizing the nonlinear system.
2. **Unscented Kalman Filter (UKF)**: Use the unscented transformation to approximate the first two moments of the true nonlinear distribution and propagate these transformation points through the nonlinear dynamics.
3. **Ensemble Kalman Filter (EnKF)**: Randomly sample a set of particles from the state distribution and propagate these particles through the nonlinear system dynamics.
4. **Particle Filter (PF)**: Sample a set of particles from the state distribution and resample these particles according to the observed values to reflect their observed likelihood.

The main contribution of the paper lies in providing an overview and comparison of these methods and guidance for their efficient implementation. By simulating the modified four - tank system, the paper evaluates the performance of these methods, including the computation time and the accuracy of state estimation (measured by the Mean Absolute Percentage Error, MAPE).