Implement better handling of the reference values of the estimated orbital parameters in Kalman filter
In KalmanEstimator class, the orbital parameter drivers' reference values are set at the initialization of the filter and never change.
This is an issue. Indeed since the Kalman filter constantly updates its date and orbit, the reference value gets further away from the actual value and we're losing the point of performing a normalization.
For example, if the orbit at initialization is geostationary, half an orbit after initialization, the position will be ~84000km away from the reference position.
As we are using this reference value for normalization in order to avoid numerical inconsistencies... this behaviour is counterproductive.
The proposed solution is to reset the reference values of the orbital drivers when these drivers' values are reset in the propagator builder. Doing that, the normalized values of the orbital drivers in the filter will then represent the scaled difference between the corrected orbital states and their predicted counterparts. With this, the normalization process would get back its original purpose.
In concrete terms, the orbital drivers values are reset in the filter in method predictState from class Model (l. 898). The method called to reset them is method resetOrbit from class AbstractPropagatorBuilder (l.278). The change consists in updating the reference values in the same fashion the actual values are updated in this method.