ST-GMLP: A concise spatial-temporal framework based on gated multi-layer perceptron for traffic flow forecasting.

Area of Science:

• Artificial Intelligence
• Urban Planning
• Transportation Engineering

Background:

• Traffic congestion is a major urban challenge, necessitating accurate traffic flow forecasting.
• Existing models often struggle with complex temporal patterns and computational demands, limiting real-world applicability.
• Temporal distribution shifts between historical data periods hinder forecasting accuracy.

Purpose of the Study:

• To propose a concise and effective framework, Spatial-Temporal Gated Multi-Layer Perceptron (ST-GMLP), for enhanced traffic forecasting.
• To leverage multi-scale temporal patterns and spatial-temporal interdependencies for improved prediction accuracy.
• To mitigate the adverse effects of temporal distribution shifts in traffic data.

Main Methods:

• Developed ST-GMLP, a framework utilizing a Gated Multi-Layer Perceptron (GMLP) for modeling spatial-temporal dependencies.
• Employed a parallel structure to learn interdependencies in both spatial and temporal directions.
• Established interactions between time and space to address temporal distribution shifts.

Main Results:

• ST-GMLP demonstrated superior performance compared to existing state-of-the-art traffic forecasting methods.
• The model showed significant advantages in training efficiency and resource utilization.
• Experimental findings confirmed the effectiveness of ST-GMLP in handling temporal distribution shifts.

Conclusions:

• ST-GMLP offers a simple yet effective structure for accurate traffic flow forecasting.
• The framework successfully leverages spatial-temporal interdependencies to overcome limitations of traditional models.
• ST-GMLP presents a promising solution for real-world traffic management and congestion alleviation.