Evaluation and comparison of Kalman filter algorithms

Abstract

In practical applications, biogas flow sensors are used to measure the output methane gas from the reactor. But none of the states are measured. State estimation methods can be used to solve this problem. In this thesis the problem of optimal state estimation regarding to an anaerobic digestion (AD) reactor is considered. For the nonlinear system, different approaches: extended Kalman filter (EKF), unscented Kalman filter (UKF) and particle filter (PF), are used to estimate the unmeasured states. The main aim is to evaluate and compare the three alternative Kalman filter algorithms with a proper state augmentation, both in terms of estimation accuracy and computational efforts, by applying them to a real and a simulated AD reactor. Both the simulation and test with real data show that UKF has a similar performance with EKF, but more susceptible to the initial condition. PF has the best state estimate performance with the highest computational requirement, about 30 times more CPU time than the EKF. Overall we conclude that the EKF, with Jacobian calculations every time step, is the best choice for the unmeasured state estimation in this case.