Spatio-temporal characterisation and compensation method based on CNN and LSTM for residential travel data

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Traffic simulations often require extensive travel behavior data, which is challenging to collect due to privacy concerns and data volume.
• Existing methods struggle to accurately model resident travel patterns with limited datasets.

Purpose of the Study:

• To develop a method for more accurate traffic simulations using limited resident travel data.
• To leverage deep learning for analyzing and compensating spatiotemporal features in travel behavior.

Main Methods:

• A hybrid model combining Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for time-series analysis.
• Modeling spatiotemporal travel patterns to achieve realistic traffic simulations.

Main Results:

• The proposed CNN-LSTM method demonstrated superior performance compared to modulation and statistical estimation methods.
• Experimental results showed the model's output for average traveling distance was closer to real-world data.
• The strategy significantly reduced model deviation, achieving an approximate 50% reduction in the basic error rate.

Conclusions:

• The CNN-LSTM approach effectively models spatiotemporal travel data, enabling more accurate traffic simulations with limited input.
• This method offers a viable solution for overcoming data acquisition challenges in traffic simulation.
• The study highlights the potential of deep learning in enhancing the realism and accuracy of transportation modeling.