SHORT-TERM TRAFFIC FLOW PREDICTION OF SMART CITIES BASED ON SPATIAL-TEMPORAL GRAPH CONVOLUTIONAL RECURRENT NEURAL NETWORK

To enhance the accuracy of short-term traffic flow prediction for urban traffic flow prediction, a prediction model combining spatial-temporal graph convolutional network and recurrent neural network is designed.The method first models the traffic network as a graph structure, uses spatial-temporal graph convolutional network to capture the spatial relationship between traffic sensors or road segments, and extracts the features of adjacent nodes through graph convolutional layer.To enhance the processing capability for time series, a recurrent neural network structure is further introduced to learn temporal dynamic changes through recurrent units.Meanwhile, the model introduces attention mechanism to enhance its feature extraction capability.By conducting experiments on the traffic flow data of a certain working day in Chengdu, multiple models are used for comparative analysis.The MAPE of the proposed model during peak hours ranged from 0.090 to 0.092.The RMSE ranged from 17.440 to 14.687.The average absolute error ranged from 13.180 to 10.709.In the error indicators of traffic flow during off peak hours, they were 0. 068, 11.192, and 9.314.The model can effectively capture the dynamic changes of traffic flow in various time periods, thereby significantly improving the prediction accuracy and providing strong support for urban traffic management.