Researchers have focused much attention on how to improve the prediction accuracy of traffic intensity because it is well known that the construction of intelligent transportation systems (ITSs) is particularly important for future urban development and that accurate traffic prediction is the key to improving the safety and efficiency of ITSs. On the one hand, the complex spatio-temporal relationships behind traffic patterns always affect the prediction accuracy, traditional models are used to extract spatio-temporal dependence via two different neural networks separately, and this type of combined model ignores spatio-temporal coexistence. On the other hand, researchers who utilize only past traffic data gathered in the hour before the predicted time point to build a case neglect similitude in traffic data where there are obvious repeats. This study proposes a spatio-temporal unified aggregate transformer (STUAformer), which first establishes a spatio-temporal matrix to maintain cross-space-time connectivity between nodes; then, traffic speeds from previous days that coincide with the historical observation series span for the present day are included as periodic references; and finally, the transformer is employed to mine the hidden spatio-temporal features behind the traffic data. Finally, utilizing three publicly available datasets, we compare the model with several state-of-the-art models. The findings demonstrate that the model greatly enhances traffic prediction.