Orbit Determination doesn't work!

Hi, all,
These couple of weeks, I tried to use OrbitDetermination in my simulation but failed to get satisfying reults, not any result had high precision. I am definitly sure there is something wrong in my code, and I am here for some help.
First, let me introduce my simulation. I took the CPF data of Stella from July 02-03, 2012. I set a ground station, and simulated measurements by just add range/az/el noises (30m/0.1deg/0.1deg) to the RAE generated from the CPF data. I know in Orekit, a good measurement should take into account of all physical and practical effects. Therefore, in order to continue my simulation with Orekit, as you can see in my “orbit-determination.in” file, I set “twoway = false”, “range/az/el.bias=false”, “on.board.clock.offset.estimated = false” , etc., and just drag and solar effects are estimated.
orbit-determination.in (18.1 KB)
W3B.aer (972.0 KB)

The result of Orbit fit shows that the estimated drag effect is 10^3, the same to the solar effect. It is weird, isn’t it? What is worse, the extrapolation errors of x/y/z over 3 days can reach 10km. And the more data I used, the accuracy is worse.
Here is the errors of x/y/z:
errors of xyz over 5 days.docx (48.3 KB)

Does anyone have any suggestions?

RAE generated from the CPF data

How did these values get generated? The measurements data seem to not have been generated properly to me. Looking at the elevation for example, some measurements have a negative elevation (the first one in the file is 50 degrees below horizon!), so I guess you you didn’t use the measurement generation feature (using for example classes RangeBuilder, AngularAzElBuilder, EventBasedScheduler, Generator…). One typical error when not using RangeBuilder is to forget about the light time delay, so the generated measurement is inconsistent with the model used in the Range class during the orbit determination afterwards.

Another point: your measurements file involves only one ground station. You don’t get good observability when using one ground station only. You should also remove the unused stations in the orbit-determination.in file, they are misleading here.

What is worse, the extrapolation errors of x/y/z over 3 days can reach 10km

If orbit determination did not converge to the correct orbit, it is normal it diverges when you propagate a long time after the measurement time range (which covers 24 hours here).

For what we know, orbit determination does work in Orekit. It has been used for years now.