The navigation constellation's autonomous navigation requires the autonomously updating of the broadcast ephemeris. In view of the limited data-processing ability of the satellite, the forgetting factor recursive least-squares algorithm was introduced to improve the data-processing efficiency. Considering that the navigation satellite's orbit eccentricity is equal to zero approximately, it would result in the singularity of the normal matrix in the process of fitting. The new broadcast ephemeris parameters fitting algorithm was developed and the partial derivative matrix of satellite position vector with respect to modified parameter was deduced based on the non-singular transformation. Finally, the rationality of formulas and algorithm are validated by simulation.