Traffic flow prediction method based on dynamic graph neural network

Abstract

The invention relates to the field of traffic planning, in particular to a traffic flow prediction method based on a dynamic graph. The method comprises the steps of setting a plurality of monitoring sensors for key nodes of a traffic road, monitoring and collecting traffic flow, road occupancy and speed data of the road for a long time, constructing a traffic flow prediction model based on a dynamic graph neural network, inputting historical traffic flow data, setting related hyper-parameters in the model, and predicting the traffic flow. The method comprises the steps of preprocessing input data, designing a dynamic graph updating algorithm, applying the algorithm to a dynamic graph neural network module, extracting spatio-temporal features by adopting the dynamic graph neural network module and a ConvLSTM module, outputting a prediction result after the features are fused, and finally training a whole model for traffic flow prediction. According to the method, the fitting degree of the model prediction output and the actual flow condition is better, the prediction output stability is better, no large fluctuation occurs, and the traffic flow prediction result is more reliable and has more advantages.

Description

technical field [0001] The invention belongs to the field of computer / traffic planning, in particular to a traffic flow prediction method based on dynamic graphs. Background technique [0002] With the progress of urbanization in recent years, traffic problems have become increasingly serious. As a mature solution, the intelligent transportation system can carry out real-time control, real-time scheduling, and abnormal monitoring of traffic conditions, but its core cannot be separated from the real-time prediction of traffic flow. Traffic flow is an important indicator of road conditions. If the traffic flow conditions can be accurately predicted in advance, the traffic management department can guide them in a timely and reasonable manner. [0003] The traffic flow forecasting problem is a typical spatio-temporal data forecasting problem. The difficulty lies in how to extract the spatial and temporal features of the road. Specifically, spatial characteristics, that is, th...

AI Technical Summary

Problems solved by technology

However, in this process, the traffic road is abstracted into a fixed graph structure, which does not take into account that the state of the traffic road is a process of dynamic changes.

Images