The paper introduces Graph WaveNet, a novel graph neural network architecture designed for spatial-temporal graph modeling. Traditional methods often assume a fixed graph structure and struggle to capture long-range temporal dependencies, leading to limitations in handling complex system problems such as traffic speed forecasting. Graph WaveNet addresses these issues by incorporating a self-adaptive adjacency matrix and stacked dilated 1D convolution layers. The self-adaptive adjacency matrix learns hidden spatial dependencies from the data, while the stacked dilated 1D convolution layers enable the model to handle long sequences effectively. Experimental results on two public traffic network datasets, METR-LA and PEMS-BAY, demonstrate the superior performance of Graph WaveNet compared to existing methods, including ARIMA, FC-LSTM, WaveNet, DCRNN, GGRU, and STGCN. The model's effectiveness is further validated through its ability to learn hidden spatial dependencies and its efficiency in computation time.The paper introduces Graph WaveNet, a novel graph neural network architecture designed for spatial-temporal graph modeling. Traditional methods often assume a fixed graph structure and struggle to capture long-range temporal dependencies, leading to limitations in handling complex system problems such as traffic speed forecasting. Graph WaveNet addresses these issues by incorporating a self-adaptive adjacency matrix and stacked dilated 1D convolution layers. The self-adaptive adjacency matrix learns hidden spatial dependencies from the data, while the stacked dilated 1D convolution layers enable the model to handle long sequences effectively. Experimental results on two public traffic network datasets, METR-LA and PEMS-BAY, demonstrate the superior performance of Graph WaveNet compared to existing methods, including ARIMA, FC-LSTM, WaveNet, DCRNN, GGRU, and STGCN. The model's effectiveness is further validated through its ability to learn hidden spatial dependencies and its efficiency in computation time.