The construction of an integrated cloud network digital intelligence platform for rail transit based on artificial...

Area of Science:

• Intelligent transportation systems
• Construction management technology
• Artificial intelligence in engineering

Background:

• Rail transit construction faces challenges in real-time data integration and efficient scheduling.
• Existing systems often lack advanced AI capabilities for prediction and optimization.
• The need for intelligent, closed-loop control platforms is critical for modern infrastructure projects.

Purpose of the Study:

• To design and validate a closed-loop control platform for rail transit construction.
• To integrate multi-source data for real-time prediction and AI-driven scheduling.
• To leverage digital twins for strategy execution and feedback in construction management.

Main Methods:

• Developed a three-layer architecture: edge sensing, cloud computing, and intelligent interaction.
• Implemented data fusion middleware, an AI decision engine, and a 3D digital twins module.
• Utilized Transformer-Encoder, Graph Attention Networks, Apache Kafka/Flink, Spatio-Temporal Graph Convolutional Networks, Shifted Window Transformer, and Proximal Policy Optimization (PPO).

Main Results:

• Achieved 91.6% accuracy in passenger flow prediction and 98.2% in image recognition.
• Reduced average task completion time by 27.4% using PPO-based scheduling.
• Demonstrated average response latency of 280 ms, peak throughput of 27,000 messages/sec, and >95% closed-loop success rate.

Conclusions:

• The platform effectively meets design targets for prediction accuracy, latency, and scheduling efficiency.
• The system provides a robust foundation for the informatization and intelligent upgrading of urban rail transit construction.
• The closed-loop control approach enhances operational efficiency and reliability in complex construction environments.