The prediction of traffic flow can be greatly useful for the work of traffic management departments and the travel planning of drivers. How to make accurate and efficient traffic flow prediction is a very important issue. Traditional traffic flow prediction data sources are usually vehicle speed and driving trajectory. The researchers obtained data by arranging traffic sensors on the highway at regular intervals. These methods applied to suburban areas and highways have achieved good results.It is inconvenient to obtain these data on dense and complicated urban roads, and it is impossible to use existing methods to make predictions. Therefore, this paper proposes a method for forecasting by using traffic flow data of urban road checkpoints. We first analyze the characteristics of cyclical changes in traffic flow by analyzing existing traffic data. Then we extract corresponding features based on these cyclic changes. Finally we train traffic flow prediction models which are suitable for urban checkpoints based on these characteristics. We carried out a large number of experiments based on real traffic data sets, and the results show that our traffic flow prediction model has a good prediction effect, with RMSE and MAPE of the predicted values of 15.3 and 7.3, respectively，the accuracy can reach 92.7%.