A Distributed VMD-BiLSTM Model for Taxi Demand Forecasting with GPS Sensor Data.

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Taxis are crucial for public transport, but vacant vehicles represent significant resource waste.
• Short-term taxi demand forecasting balances supply and demand, reducing fuel consumption.
• Existing machine learning and deep learning models struggle with computational costs and large-scale trajectory data.

Purpose of the Study:

• To develop an efficient hybrid deep learning model for accurate taxi demand prediction.
• To address the computational challenges of existing models using large-scale sensor data.
• To improve taxi operational efficiency and profitability.

Main Methods:

• A hybrid model combining Variational Mode Decomposition (VMD) and Bidirectional Long Short-Term Memory (BiLSTM).
• VMD decomposes traffic features into frequency-specific sub-modes.
• BiLSTM predicts demand using decomposed features, optimized for performance on the Spark distributed platform.

Main Results:

• The proposed VMD-BiLSTM model demonstrates superior accuracy and efficiency compared to state-of-the-art methods.
• The model effectively utilizes large-scale taxi trajectory data.
• Performance was validated on a real-world dataset, showing significant improvements in distributed computing.

Conclusions:

• The hybrid VMD-BiLSTM model offers an effective solution for taxi demand prediction.
• This approach enhances passenger search efficiency and increases taxi profitability.
• The model's distributed performance makes it suitable for large-scale transportation networks.