<p>Accurate traffic flow prediction is crucial to improving the safety and efficiency of transportation systems. Although existing deep learning methods have made some progress in traffic flow prediction, most methods fail to effectively capture multi-source spatial relationships and long-term temporal dependencies. To address these problems, we present a traffic prediction model based on convolutional trend-aware attention and adaptive spatio-temporal graph convolution (TA-ASTGCN). The model captures complex spatio-temporal relationships by alternately stacking multiple convolutional trend-aware attention layers and adaptive spatio-temporal graph convolution layers in the encoder and decoder. The convolutional trend-aware attention layer captures the trends in traffic flow sequences and the correlations between nodes. The adaptive spatio-temporal graph convolution layer adaptively adjusts the strength of node relationships to effectively mine spatial correlations in traffic data. Compared with ASTGNN, with the best predictive performance in the baseline models, our TA-ASTGCN showed on average 3.69%, 3.14%, and 3.80% improvement in MAE, RMSE, and MAPE on four datasets. Experimental results showed that the prediction performance of the proposed model is better than the baseline models.</p>