Trying to propagate orbits with more than 200 (grade and order) using SH is very slow, is it possible to replicate the SH model using a neural network or something similar achieving faster propagations?

  • $\begingroup$ I have a hunch this is an interesting question, but what is "a ML model"? Implementing a neural network is hardware intensive, it would probably be faster much to use that same hardware to implement a spherical harmonic evaluator directly, but that's just my guess. Neural networks are not known for a dozen digits of numerical accuracy, since they are often based on few or one-byte integer math, so I think they are the wrong tool for the job of orbit propagation, but maybe there are arguments to the contrary. $\endgroup$
    – uhoh
    Sep 4 at 1:44
  • 1
    $\begingroup$ @uhoh "ML model" in this case, would just be "machine learning model". I've suggested the edit. $\endgroup$ Sep 4 at 1:57
  • 2
    $\begingroup$ In most cases this would ”work” but give worse results and taking longer to compute. This does not sound like the type of problems ML is good at. $\endgroup$
    – lijat
    Sep 4 at 5:29
  • $\begingroup$ Thank you for your response, I agree with both of you (@uhoh, @lijat), but I think there should be another way to achieve faster computation times, I did try mascon models but they are not so great since in the end parallelization works better with "intensive and long" tasks, maybe polyhedral models could work. Is there another way to replicate SH that should be tried? $\endgroup$ Sep 4 at 17:38
  • $\begingroup$ One solution to the slowness of a 200 degree and order spherical harmonic model is not to use a 200 degree and order spherical harmonic model. Truncate the model -- and use one of the spherical harmonics algorithms. You'll get a better and faster approximation than a ML model. $\endgroup$ Sep 5 at 12:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.