Origin-destination prediction from road average speed data using GraphResLSTM model

Area of Science:

• Intelligent Transportation Systems (ITS)
• Traffic Management
• Data Science

Background:

• Accurate origin-destination (OD) prediction is vital for traffic management and resource allocation in ITS.
• Traditional methods often rely on traffic flow data, which may not capture congestion nuances effectively.

Purpose of the Study:

• To present a novel integrated framework, GraphResLSTM, for enhanced OD prediction.
• To evaluate the efficacy of road average speed data against traditional traffic flow data for OD prediction.

Main Methods:

• Developed GraphResLSTM by integrating Graph Convolutional Network (GCN), Residual Neural Network (ResNet), and Long Short-Term Memory (LSTM).
• Utilized road average speed data, simulated using Simulation of Urban Mobility (SUMO) on a real-world road network.
• Employed a hybrid entropy weight method and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) for key road segment selection to improve training efficiency.

Main Results:

• The GraphResLSTM model demonstrated superior performance in OD prediction.
• Road average speed data proved more effective than traditional traffic flow data for OD prediction.
• Comparative experiments confirmed the model's and data type's advantages over alternatives.

Conclusions:

• The GraphResLSTM framework offers a significant advancement in OD prediction accuracy within ITS.
• Road average speed data provides richer insights for traffic prediction compared to flow data.
• The proposed methods enhance both prediction accuracy and training efficiency.