Multitask Learning and GCN-Based Taxi Demand Prediction for a Traffic Road Network.

Area of Science:

• Intelligent transportation systems
• Deep learning applications in urban mobility
• Traffic flow analysis and prediction

Background:

• Accurate urban taxi demand forecasting is crucial for intelligent transportation systems.
• Forecasting is challenging due to complex spatial-temporal dependencies, dynamic traffic, and uncertainty.
• Existing methods struggle to fully utilize global and local correlations in traffic flow.

Purpose of the Study:

• To develop a novel deep learning model for enhanced urban taxi demand prediction.
• To effectively capture spatial-temporal correlations in taxi trip data.
• To improve the generalizability and accuracy of traffic flow forecasting.

Main Methods:

• Utilized a graph convolutional network (GCN) to model spatial patterns of taxi trips on road networks.
• Employed long short-term memory (LSTM) networks to extract temporal features of traffic flows.
• Implemented a multitask learning strategy to enhance model generalizability and performance.

Main Results:

• The proposed model demonstrated high efficiency and accuracy in real-world taxi trajectory data experiments.
• The model effectively forecasts short-term taxi demands at the traffic network level.
• Experimental results indicate superior performance compared to state-of-the-art traffic prediction methods.

Conclusions:

• The deep learning model integrating GCN, LSTM, and multitask learning is effective for urban taxi demand forecasting.
• The approach successfully addresses the complexities of spatial-temporal dependencies in traffic data.
• This method offers a significant advancement in intelligent transportation research for accurate traffic prediction.