After reading the article 'Application of the unscented Kalman filter for real-time nonlinear structural system identification',the author used the EKF and UKF to identify the parameters of a two degree linear structural system under the earthquake excitation, and the EKF and UKF both have good convergent results. But when I use the EKF to identify parameters of a four degree linear system, it doesn't have good results. Here is the Matlab order, would you please have a look at it? It is my honor if I could everybody's help and give me some suggestions.And SSI is the main order. Thank you very much!
Dear Yun Wang,
First:
- Precautions are taken to ensure that the data represented in this area are consistent.
Secondly
- Initially the UKF filter has been developed to treat highly nonlinear case avoiding the linearization phase made in the EKF. the difference between the approach EKF and UKF approach for data integration. However it may present in certain conditions a processing delay and becomes less efficient than the EKF. Be careful!
Thirdly here 2 attached files to study.
Best regards
What are your estimated states? If there are angles involved, you may have to take a few specific steps to make sure that the UKF mean and covariance are computed correctly (in fact, this is true for any periodic spaces). Look for papers by J. Crassidis on the UKF and attitude estimation.
If you are estimating parameters (constant but unknown values), add a small amount of process noise to improve filter convergence.
Also, system input must be sufficiently "rich" to ensure convergence of parameter estimates to their true values (for example, you can't estimate mass from acceleration-free motion).
- Badges
- Science topic