Advanced robotic applications have revived interest in identification of a high-precision dynamic model. In this article, we propose an extended dynamic parameter set (EDS). The EDS breaks through the limitation that the base dynamic parameter set needs a priori knowledge of the gravity direction for modeling. Moreover, we present a novel parameters identification technique (RSIH), which is a complete solution and can significantly mitigate negative effects of the measurement noise and outliers. Besides, an incremental learning technique combined with a compensation limit criterion is employed to compensate for unmodeled dynamics. Simulations and experiments demonstrate the EDS-based model can adapt to any installation angle of a base plate, and confirm the RSIH technique outperforms the widely used identification techniques in industry and is equal to or even better than the state-of-the-art physical feasibility technique in terms of identification precision and robustness. In addition, the modeling errors, especially the uncertainty of the friction model, can be greatly compensated.