Traffic flow prediction is always considered to be a challenging task due to the complex periodicity of traffic flow data, the dynamic nature of real-time change and the spatial-temporal dependence of road network. Aiming at the above problems, this paper proposes a new traffic flow prediction model based on Extended Multi-component Graph Convolutional Network (EMGCN), which solves the problem that the existing traffic flow prediction methods are not sufficient to extract spatial-temporal features. Specifically, we introduce extended multi-component module to extract periodic temporal diffusion information of traffic flow. Then, the encoder consists of gated recurrent unit and graph convolutional network to capture the spatial-temporal features of traffic flow, and finally converts the spatial-temporal features into future prediction sequences through the decoder. Extensive experiments are carried out on the public traffic datasets PEMSD8, and the experimental results show that the proposed model achieves the best results in all aspects.