It is impossible to overestimate the importance of knowing(or getting or Estimating) a parameter at the control of the system, monitoring and a prediction of view. However, it’s so complicated to estimate a parameter because of the non-linearity in the dynamic system. In this paper, the parameter estimation is handled as a requirement to improve the control stability in the dynamic system.
There are methods to estimate parameter which are Least Square Method (LSM) and Maximum Likelihood Method (MLM). However, the electronic system and mechanical system are rapidly developed in modern society. By those reasons, Unscented Kalman Filter (UKF) which is non-linear filter is adopt to estimate the status and parameters of dynamic system in non-linear system. Unscented Kalman Filter (UKF) has a strong point that does not generate errors of the linearization using the Unscented Transformation (UT). Moreover, Extended Kalman Filter (EKF) and UKF are widely used in the localization of a robot, trajectory estimation, weight learning of the neural network.
Therefore, EKF and LSM are selected for the comparative research. The performance of these methods is verified through the estimated results.
In this paper, the experimental environment that composed of Coupled Water Tank system and 2-Degree of Freedom pendulum system for the dynamic systems is designed to verify the estimated performance. Furthermore, I created simulation data from the mathematical model of the dynamic systems and analyzed the performance of using the exact data. In order for professional analysis, I implemented “monte carlo simulation” 100 times for each simulation and I organized into graphs and tables.
The other analysis is robustness test about measurement noise on these algorithms.
As a result, it is confirmed that the performance of UKF is better than EKF and LSM.