Optimization of two-passenger ride-pooling orders based on ST-GNN and path optimization.

Area of Science:

• Intelligent Transportation Systems
• Urban Mobility Optimization
• Data Science

Background:

• Urban ride-pooling faces challenges in real-time order matching and path planning due to complex spatio-temporal demand and traffic.
• Existing algorithms struggle to integrate these dynamic features for multi-objective optimization under real-world constraints.

Purpose of the Study:

• To propose a novel dual-optimization framework for efficient urban dynamic ride-pooling.
• To enhance real-time order matching and path planning by integrating spatio-temporal dynamics and multi-objective optimization.

Main Methods:

• Constructed a demand-adaptive urban spatial structure using Voronoi polygons.
• Developed a Spatio-Temporal Graph Neural Network (ST-GNN) with multi-head attention and Transformer mechanisms to learn urban dynamics.
• Integrated ST-GNN embeddings with an improved Dijkstra algorithm for multi-objective path planning (distance, detour, emissions).

Main Results:

• Achieved an 86.6% matching success rate on a large-scale real-world dataset.
• Reduced average carbon emissions by 0.34 kg CO2 per order.
• Maintained a low average detour rate of 0.1202.

Conclusions:

• The proposed framework effectively enhances spatio-temporal collaboration in complex ride-pooling scenarios.
• Offers a practical and efficient solution for intelligent shared mobility systems, promoting optimized urban traffic and low-carbon travel.