 Combined SAI-SHAO solution |
90-day forecast for X, Y, UT1-UTC.
Combination is based on four methods
1) SAI AR, 2) SAI LSC, 3) SAI NN 4) SHAO LS+AR.
Weights are calculated from MJD 55659-55900 statistics
files can be downloaded from
http://lnfm1.sai.msu.ru/~tempus/pvz/prediction/saishao/
in the format saishaoMMMMM.dat
where MMMMM is 5-digit MJD
|
|
|
Autoregression method |
Autoregression of the 50-order, estimated by Burg algorithm 6-year base period (~2190 points).For UT1-UTC and LOD tidal variations in the Earth rotation IERS Convention 2000 model excluded, then included and predictions are performed for differences. Trend of the second order modeled for all, except LOD. Annual harmonic fitted for X, Y, UT1UTC, Chandler - for X,Y, semiannual for UT1UTC by nonlinear LSM (amplitude, phase, period). |
|
|
Least Squares Collocation method |
Autocorrelation function estimated from 6-year base period (~2190 points), last 200 are points used for prediction.
For UT1-UTC and LOD predictions are performed for differences. Trend of the second order modeled for all, except LOD. |
|
|
Neural network |
3-layered neural network with 15 neurons;
from the 6-year base period (~2190 points) 100-pints training vectors are formed, and the network learns to predict the next points by
Levenberg-Marquardt backpropagation study method; after 50 epochs of study from the last 100 points the next point is predicted and recurrently 60 points of prediction are formed For UT1-UTC and LOD predictions are performed for differences. |
Initial data used: eopc04_IAU2000.62-now
Made for Earth Orientation Parameters Combination of Prediction Pilot Project