Spatiotemporal prediction is widely used in the fields of neuroscience, climate, and transportation, and traffic speed prediction is one of the typical research areas. Since traffic networks are irregular grid structures with complex nonlinear spatiotemporal dependencies between nodes, traditional single-feature prediction methods are difficult to adapt to complex road conditions. In order to improve the prediction accuracy of traffic speed, this paper proposes a spatiotemporal traffic speed prediction method combining attention and multivariate graph convolution fusion (CAMGCF). It constructs a spatiotemporal model, specifically, the network uses multivariate graph convolution to fuse external factor features and traffic speed features and uses an attention mechanism to adaptively fuse feature information from different time series to achieve stability in long-term prediction. Finally, the framework is evaluated on three real-world large-scale road network traffic datasets, and the experimental results show that the prediction accuracy of this paper has more accurate results compared with the benchmark method.