Accurate and reliable traffic speed prediction is crucial for enhancing traffic management efficiency and alleviating traffic congestion. To improve the traffic speed prediction accuracy and capture complex dynamic spatio-temporal dependencies in traffic data, this study proposes a traffic speed prediction model based on the multi-scale dynamic spatio-temporal convolutional network (MDSTCN). Firstly, an adaptive static spacial adjacency matrix is constructed based on the graph convolutional network, and attention mechanism is introduced to capture the directional dependencies and dynamic interactions of traffic flow, thereby uncovering the intricate dynamic spatial relationships. Secondly, a multi-scale causal dilated convolutional structure is adopted to extract temporal correlations encompassing both local patterns and long-term trends. Finally, the model is trained and tested on a real-world dataset, with a variety of comparative experiments designed. The results show that compared with the best baseline model Graph WaveNet, the proposed MDSTCN model achieves average relative reductions of 5.91% in RMSE, 7.51% in MAE, and 8.59% in MAPE. Furthermore, the prediction error fluctuation across 15/30/60-minute time horizons is minimal, highlighting its good adaptability and stability in long-term prediction tasks.