Traffic Congestion Prediction on Urban Roads Based on Multiple Types of External Datasets

With the acceleration of urbanization and the increasing demand for human travel, the problem of traffic congestion on urban roads has become increasingly serious.Strengthening the construction of Intelligent Transportation System (ITS) and improving the accuracy of traffic flow prediction are the core issues to slow down traffic congestion and traffic accidents.This study selects the section of Xizhimen North Avenue in Beijing as the research object, introduces external variables such as subway station passenger flow, holiday information and meteorological data, and adopts four models, namely, SARIMA, SVR, GRU and LSTM, for traffic flow prediction.The results show that the deep learning-based LSTM and GRU models are excellent in prediction performance, with R values over 0.79 and MAE and MSE values less than 0.1, which are significantly better than SVR and SARIMA models.These results validate the important influence of external factors on traffic flow and demonstrate the potential of deep learning methods in traffic prediction, which provide a scientific decision-making basis for the traffic management department and help to formulate more effective traffic diversion and management strategies to alleviate the urban traffic pressure.