This paper proposes a novel graph neural network architecture called Graph WaveNet for spatial-temporal graph modeling. The model addresses the limitations of existing approaches by capturing both spatial and temporal dependencies. It introduces a self-adaptive adjacency matrix that learns hidden spatial dependencies from data without prior knowledge. The model also uses stacked dilated causal convolutions to efficiently capture long-range temporal sequences. The framework integrates these components seamlessly in an end-to-end manner. Experimental results on two public traffic datasets, METR-LA and PEMS-BAY, demonstrate that Graph WaveNet achieves state-of-the-art performance. The model outperforms existing methods in both spatial and temporal modeling, and is efficient in terms of computation. The self-adaptive adjacency matrix is shown to be effective in capturing hidden spatial dependencies, even when the graph structure is unavailable. The model is also efficient in terms of computation, with faster training and inference times compared to other methods. The paper concludes that Graph WaveNet is a promising approach for spatial-temporal graph modeling, and future work will focus on applying it to large-scale datasets and learning dynamic spatial dependencies.This paper proposes a novel graph neural network architecture called Graph WaveNet for spatial-temporal graph modeling. The model addresses the limitations of existing approaches by capturing both spatial and temporal dependencies. It introduces a self-adaptive adjacency matrix that learns hidden spatial dependencies from data without prior knowledge. The model also uses stacked dilated causal convolutions to efficiently capture long-range temporal sequences. The framework integrates these components seamlessly in an end-to-end manner. Experimental results on two public traffic datasets, METR-LA and PEMS-BAY, demonstrate that Graph WaveNet achieves state-of-the-art performance. The model outperforms existing methods in both spatial and temporal modeling, and is efficient in terms of computation. The self-adaptive adjacency matrix is shown to be effective in capturing hidden spatial dependencies, even when the graph structure is unavailable. The model is also efficient in terms of computation, with faster training and inference times compared to other methods. The paper concludes that Graph WaveNet is a promising approach for spatial-temporal graph modeling, and future work will focus on applying it to large-scale datasets and learning dynamic spatial dependencies.