The　diagnosis　of　epilepsy　in　hospital　is　mostly　judged　by　experienced　medical　personnel　visually　observing　brain　waves　combined　with　some　characteristic　clinical　manifestations.　As　brain　signal＇s　complexity　the　understanding　of　EEG　signal　still　remains　challenge.　In　this　paper　we　proposed　a　novel　seizure　prediction　method　based　on　generative　features.　Then　predict　seizures　according　to　the　EEG　information　of　the　epileptic　patients.　And　we　use　the　Extreme　learning　machine　as　the　classifier　of　generative　features.　Finally,　we　get　the　highest　accuracy　score　of　98%.