Traffic Flow Prediction Model Based on the Combination of Improved Gated Recurrent Unit and Graph Convolutional...

Area of Science:

• Intelligent Transportation Systems
• Traffic Engineering
• Data Science

Background:

• Increasing vehicle numbers lead to severe traffic congestion, necessitating advanced traffic management solutions.
• Effective traffic flow prediction is challenging due to complex road networks and population mobility.

Purpose of the Study:

• To develop an accurate traffic flow forecasting model for Wenyi Road in Hangzhou.
• To address the spatiotemporal correlations inherent in traffic flow data.

Main Methods:

• Proposed the IMgru model to capture temporal traffic flow characteristics.
• Developed the IMgruGcn model, integrating Graph Convolutional Networks (GCN) with IMgru for spatiotemporal feature extraction.
• Segmented the Wenyi Road dataset into peak and off-peak periods for targeted prediction.

Main Results:

• The IMgruGcn model demonstrated superior performance compared to five baseline models and a state-of-the-art method.
• Achieved optimal results on the Wenyi Road dataset and a public dataset, confirming its generalization capabilities.

Conclusions:

• The IMgruGcn model effectively extracts spatiotemporal traffic flow features.
• The proposed model offers a robust solution for traffic flow prediction, applicable to real-world intelligent transportation systems.