I am working on implementing a Kalman filter integrated with ARMA parameters, as described in the article "Predicting Time Series Using an Automatic New Algorithm of the Kalman Filter" (Mathematics, 2022). In the proposed method, the authors assume no observation error, effectively setting the observation variance to zero. However, when I applied this approach using the pykalman library in Python, the estimated values matched the real data exactly, which appears unrealistic and not practical.
No, it is not logical to have the observation variance equal to 0 in a Kalman filter, as it implies perfect measurement accuracy without any noise. In practice, some level of observation noise is always present, so the observation variance is typically greater than 0.
Indeed, setting the observation error to zero in uneralistic. The Kalman Filter will always be equal to the measurement value with no uncertainty. Normally setting the observation covariance to a small value would help the filter not be overconfident
- Badges
- Science topic
- Similar topics
- Filing