Most of the traffic flow sequences were single-step prediction. To realize multi-step prediction of traffic flow sequence, a convolutional neural networks (CNN)-gate recurrent unit (GRU) model based on encoder-decoder (ED) framework was proposed, abbreviated as ED CNN-GRU model. Firstly, CNN was used as en coder to extract the features of traffic flow sequence. Then GRU was used as decoder to interpret and output the extracted features. Experiments show that compared with CNN and GRU models, ED framework effectively solves the problem of rapid accumulation of errors. Compared with other benchmark models, CNN and GRU models are superior in feature extraction and interpretation of traffic flow series. For the traffic flow prediction task of 12 steps in the future, compared with other benchmark models, the root mean square error of the univariate input ED CNN-GRU model is reduced by about 0.344~6.464, and the mean absolute error is reduced by about 0.192~0.425. Compared with univariate input, the ED CNN-GRU model with multivariate input has a better fitting effect. It is proved that ED CNN-GRU model has good forecasting ability in multi-step traffic flow forecasting tasks with different input dimensions, and provides a multi-step traffic flow forecasting model supporting univariate and multivariate input for cities with different data acquisition conditions.