Research On Urban Traffic Flow Prediction Model And Method Based On Data Fusion

With the expansion of urban scale and infrastructure,urban roads are becoming more and more complex,and the number of vehicles on urban roads is also increasing,which brings a series of traffic problems such as traffic safety and congestion.Intelligent Transportation Systems(ITS)is an effective solution currently used,and traffic flow prediction has become an important research hotspot in the key technologies of ITS.Traffic flow forecasting can provide traffic data support and suggestions for urban management systems.It can also provide reliable traffic forecasting reports and plan travel routes for travelers,thereby saving travelers’time and cost costs and improving the traffic efficiency of the entire system.Traffic flow prediction is a basic and challenging task in the field of intelligent transportation.In order to improve the accuracy of traffic flow prediction,it is not only necessary to use real-time traffic flow information and historical traffic flow information,but also to consider the influence of various external factors.In recent years,graph neural network model has become the main research method of traffic flow prediction,and some significant progress has been made.Among them,graph convolutional network can effectively model the spatial dependence of traffic data,and recurrent neural network can effectively model the time dependence of traffic data.In addition to spatiotemporal factors,the impact of external factors on traffic has also been widely studied.Some scholars add weather data or Point of Interest(POIs)data around nodes as influencing factors to the transportation network,while others use knowledge graph to integrate knowledge information in the transportation network.However,due to the heterogeneity of data and inappropriate fusion methods,it is often difficult to improve the prediction effect when integrating external factors,and may even reduce the performance of the prediction network.This paper proposes two traffic flow prediction models based on data fusion to solve the above problems.The following research results have been achieved:(1)A temporal graph convolutional traffic flow prediction model based on data fusion(DFTGCN)is proposed.The existing models rarely consider external information that is not directly related to traffic,and simply incorporating external information cannot improve the prediction accuracy.The network embedding algorithm is adopted.Through the spatial feature embedding and the temporal feature embedding,the data that is not directly related to the traffic flow is added to the network,and then the attribute information of the node and the corresponding traffic characteristics are integrated to enhance the model’s ability to perceive information,thereby improving the performance of the traffic flow prediction model.(2)A graph convolutional gated recurrent network traffic flow prediction model based on data fusion(DFGCGRN)is proposed.It mainly aims at the limitations of simple models in extracting long-term data dependence and failing to consider the impact of external traffic information.The DFGCGRN model uses a encoder-decoder architecture to capture long-term dependencies in time series and handle global dependencies.At the same time,external information is integrated into the network through network embedding to enhance model input,thereby improving the performance of the model in multi-step prediction.In order to verify the effectiveness of the model,this study conducts simulation experiments on real data sets and compares it with other traffic flow prediction methods,using RMSE,MAE,ACC and R2 four performance indicators as the basis.The experimental results show that compared with other methods,embedding and fusing external information is effective in improving the performance of traffic flow prediction model,and the four performance indicators are optimized by at least 1.54%,1.67%,0.42%and 0.35%in the shortest time step prediction.