Orekit can model maneuvers in two different ways:
- Impulse maneuvers using event detectors
- Finite maneuvers using force models
For your test case, you must consider the second point.
The force models for maneuvers are in the org.orekit.forces.maneuvers package. The architecture of the package is well presented by @MaximeJ in the following forum thread. You can also find information about the maneuvers package in the technical documentation of Orekit.
To initialize the Java objects modelling maneuvers, you have different examples in different place. First, you can look at the tests classes. Then, you can look at the different tutorials:
-
ApogeeManeuver : model an apogee maneuver with a constant thrust.
-
NumericalOrbitDetermination, specifically the method setManeuver()
If there is an unmodeled maneuver in the first 24H, I am not surprised at the negative value of the reflection coefficient. The coefficient absorbs unmodeled dynamic effects and its estimated value is not physical.
I remember an Orekit validation test using the W3B satellite. The particularity of this satellite is that its life ended because of a leak. At the beginning, Orekit had not modeled the leak and the estimated reflection coefficient was therefore about -1000 
Now Orekit can model the leak, using an empirical acceleration, and the reflection coefficient is well estimated.
Best regards,
Bryan