An improved spatio-temporal graph convolution gated neural network with multi-information fusion of distance and periodic factors was proposed, in order to further explore the spatio-temporal characteristics of traffic flow and improve the prediction accuracy. Each section of the road network was taken as the node and the road network diagram structure was established according to the adjacency relationship between the sections. Considering the influence of distance between sections on spatial correlation, the spatial influence degree between sections was calculated, and different weight values were given to the adjacency matrix of the graph. In the model construction, space graph convolution module and time series prediction module were designed. The spatial feature information was extracted by the spatial graph convolution module, and the extracted spatial feature information was fused with the traffic flow cycle sequence information, and then introduced into the time series prediction module. The gating mechanism algorithm of gated recurrent unit (GRU) was redesigned to achieve the purpose of multi-source information input, and finally the predictive output was obtained. The real highway traffic flow PEMS data set was used for multi-period testing. Experimental results show that the prediction error of the proposed model is lower and the prediction performance is better than that of the current prediction method based on graph convolution.