In order to solve the problem that the traditional comfort evaluation method relies on subjective questionnaires to meet the needs of real-time and objectivity, a comfort prediction method integrating variational inference and AutoFormer+ was proposed to realize the real-time and objective evaluation of driver"s comfort. The comfort prediction method based on variational inference takes the EEG signal and sitting joint angle as inputs, and the subjective score of sitting comfort as the label, and the AutoFormer+ algorithm optimizes the attention module by introducing the frequency domain enhancement module to complete the comfort classification of the driver"s EEG data. In this paper, experiments are carried out on the static test dataset of real vehicles, and the proposed algorithm achieves better performance than the traditional method in the comfort classification task, with an accuracy of 0.95, which is 0.5% higher than that of the baseline model.