The short-term disturbance of galactic cosmic rays(GCRs) caused by coronal cass ejections (CMEs) can be quantified to indicators of geomagnetic storms. We proposed a model takes GCRs time series slice as input and generates two geomagnetic storm prediction indexes as output. This new model is based on Hilbert-Huang Transform (HHT), utilizing Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) instead of Empirical Mode Decomposition (EMD), and extends 2D spectrum result of HHT to 3D form.Two prediction indexes is constructed from the sensitive components in GCRs. Finally, this model is tested on 11 great geomagnetic storms(Kp>=8) during solar cycle 23 and 24. Accuracy of this model turned out to be 82%, and prediction lead time ranges from 9 to 24 hours.