CNN-GRU-AM for Shared Bicycles Demand Forecasting.

Area of Science:

• Data Science
• Artificial Intelligence
• Transportation Systems

Background:

• Shared bicycle demand forecasting is crucial for operational efficiency and resource management.
• Existing models often fail due to limited feature consideration and inability to capture complex temporal dynamics.
• Inaccurate forecasts lead to suboptimal vehicle distribution and reduced profitability.

Purpose of the Study:

• To develop an advanced prediction model for shared bicycle demand that overcomes limitations of existing methods.
• To enhance the accuracy of demand forecasting by integrating multiple data sources and advanced deep learning techniques.
• To improve the utilization rate of shared bicycles and project operational benefits.

Main Methods:

• A novel Convolutional Recurrent Neural Network with Attention Mechanism (CNN-GRU-AM) model was developed.
• The model utilizes a two-layer Convolutional Neural Network (CNN) for local feature extraction.
• Gated Recurrent Unit (GRU) captures temporal dependencies, and an Attention Mechanism (AM) weighs feature importance.

Main Results:

• The proposed CNN-GRU-AM model demonstrated superior prediction performance compared to other models.
• Experiments were conducted on real-world mobile and public shared bicycle datasets.
• The model effectively captures time dependence and external factors influencing shared bicycle usage.

Conclusions:

• The CNN-GRU-AM model offers a significant improvement in shared bicycle demand forecasting accuracy.
• Accurate forecasting can lead to reduced unnecessary deliveries and optimized resource allocation.
• The enhanced prediction capabilities highlight the model's potential for substantial social and economic benefits in urban mobility.