This invention discloses a method and
system for
orbit determination of BeiDou satellites by compensating for solar
radiation pressure using a
deep learning network. Addressing the problem of limited
orbit determination accuracy due to inaccurate
estimation of solar
radiation pressure parameters under regional observation networks, this invention acquires raw
satellite observation data and
space weather data to construct a dataset containing features such as
satellite position, velocity, acceleration, attitude,
solar altitude angle, and geomagnetic index. A
gradient boosting tree
algorithm is used to filter key features; a
Transformer network containing only the
encoder is used to predict
radiation pressure acceleration errors; a virtual observation equation is constructed based on the prediction error and jointly solved with the GNSS observation equation to constrain ECOM
radiation pressure parameter
estimation; and residual detection is introduced to eliminate
outlier predicted values. This invention significantly improves the
orbit determination accuracy of BeiDou satellites under regional
station networks without requiring additional hardware, and exhibits stable performance across different regions and with different numbers of stations.