Travel demand and distance analysis for free-floating car sharing based on deep learning method

Area of Science:

• Transportation Science
• Data Science
• Artificial Intelligence

Background:

• Free-floating car sharing services face challenges with temporal usage patterns.
• Understanding and predicting car-sharing demand is crucial for operational efficiency.
• Existing statistical models may not fully capture complex temporal dependencies.

Purpose of the Study:

• To develop a comprehensive time-series method for analyzing and predicting free-floating car-sharing usage patterns.
• To investigate the influence of temporal and spatial factors on car-sharing demand.
• To evaluate the performance of deep learning models against traditional statistical methods for this task.

Main Methods:

• Analysis of car-sharing booking data (e.g., car2go in Seattle).
• Application of deep learning, specifically the long-short-term memory recurrent neural network (LSTM-RNN).
• Comparative analysis with statistical models: Support Vector Regression (SVR), ARIMA, and exponential smoothing.

Main Results:

• Identified a distinct doublet pattern in car-sharing usage over time.
• Demonstrated a clear dependence of usage on population density.
• The LSTM-RNN model exhibited superior performance in statistical analysis and precision for short-term traffic characteristics like travel demand and distance.

Conclusions:

• Deep learning, particularly LSTM-RNN, offers a powerful approach to model complex temporal dynamics in car-sharing systems.
• The proposed method provides enhanced accuracy for predicting short-term travel demand and distance.
• This research contributes to optimizing the management of free-floating car-sharing services.