Aiming at the problems that different visual stimulation modulations methods will lead to low classification accuracies of some subjects, this paper designed four stimulation waveforms with four frequencies which were used to design the evoking paradigm, and the inverted sawtooth wave evoking paradigm was proposed for the first time. Eight subjects' EEG signals were collected in the experiment. By extracting and classifying frequency energy features, it was found that different stimuli have different influences on the classification accuracy. On this basis, the waveforms with the highest energy were selected to form a customized paradigm; the average classification accuracies among different stimulation waveforms and customized paradigm were compared. The experimental results show that the inverted sawtooth stimulation paradigm is better than any other traditional stimulation paradigm. Meanwhile, the average accuracy of the customized paradigm is 3%~12% higher than that of any other stimulation paradigm. Therefore, the stimulation paradigms of inverted sawtooth and customized waveforms can improve the performance of SSVEP-based BCI system.