Using a conclusion of time-constant-attenuation after loss-of-excitaion only decided by generator parameters and failure-modes, the paper according to the classify of loss-of-excitation faults, adopts a three-lay feed-forward neural network to obtain“T_q”,Which makes microcomputer-based Loss-of-Excitation protection in the twinking of fault happening failure-mode-recognition and loss-degree-prediction. Figure simulating, trend imitate testing and action moving proved that the method raised is correct.