Convolutional Long-Short Term Memory Network with Multi-Head Attention Mechanism for Traffic Flow Prediction

Area of Science:

• Transportation Engineering
• Data Science
• Artificial Intelligence

Background:

• Accurate traffic flow prediction is crucial for transportation management and decision-making.
• Data-driven methods are increasingly used for traffic flow prediction due to advancements in sensing technology.
• Existing methods struggle to capture high-dimensional feature correlations and utilize relevant data segments.

Purpose of the Study:

• To propose a novel deep learning model for improved traffic flow prediction.
• To address limitations in existing data-driven methods regarding feature correlation and data relevance.
• To enhance the accuracy and efficiency of traffic flow forecasting.

Main Methods:

• A decoder convolutional Long Short-Term Memory (LSTM) network was developed.
• Convolutional operations were employed to capture high-dimensional feature correlations.
• A multi-head attention mechanism was integrated to focus on the most relevant traffic data.
• The LSTM network was utilized to model the temporal dependencies in traffic flow data.

Main Results:

• The proposed model effectively considers correlations among high-dimensional features.
• The multi-head attention mechanism successfully identifies and utilizes the most pertinent traffic data.
• The convolutional LSTM network accurately captures temporal patterns in traffic flow.
• Experimental results on the Caltrans PeMS dataset demonstrate the model's effectiveness.

Conclusions:

• The proposed decoder convolutional LSTM network with multi-head attention significantly improves traffic flow prediction accuracy.
• This approach overcomes limitations of previous data-driven methods by integrating spatial-temporal feature correlation and data relevance.
• The findings offer a more robust solution for intelligent transportation systems and traffic management.