 This paper proposes a novel approach for predicting traffic parameters using a spatial temporal self-attention graph, convolutional network, STAGCN, model. The model uses time cosine decomposition and one-hot encoding techniques to capture the periodicity and heterogeneity of traffic flow changes. Additionally, it incorporates self-attention mechanisms into the spatial temporal convolution to capture the spatial temporal dynamic characteristics of traffic flow. This allows for improved performance compared to other baseline models, with a reduction in prediction errors of up to 5% in long-term predictions. Furthermore, the interpretability and robustness of the model are addressed by considering the spatial dynamic changes. This article is authored by Ji Hong-Chang, Chen Qinglu, and Jianmin Jia.