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

This study presents the design and validation of a closed-loop control platform for rail transit construction. The platform integrates multi-source data, enables real-time prediction, and supports AI-driven scheduling, with strategy execution and feedback implemented via digital twins. A three-layer architecture is constructed, comprising edge sensing, cloud computing, and intelligent interaction. The system incorporates data fusion middleware, an AI decision engine, and a 3D digital twins module. The operational workflow follows the perception-fusion-prediction/optimization-execution/feedback loop: edge devices collect on-site status, cloud middleware integrates and serves the data, the AI engine performs prediction and scheduling optimization, and the digital twins layer validates strategies and dispatches execution to the front end. At the data modeling level, a Transformer-Encoder-based multimodal temporal fusion model is designed, and graph attention networks are employed for heterogeneous structure modeling. Apache Kafka and Flink handle streaming data to achieve high-frequency, low-latency processing. The intelligent analysis layer integrates a Spatio-Temporal Graph Convolutional Network for passenger flow and construction period prediction, a Shifted Window Transformer for image recognition, and the Proximal Policy Optimization (PPO) algorithm for task scheduling optimization. Field tests in an urban rail construction project show that the platform maintains 91.6% accuracy in passenger flow prediction under high-concurrency conditions and achieves 98.2% accuracy in image recognition. PPO-based scheduling reduces average task completion time by 27.4%. The system sustains an average response latency of 280 ms, peak throughput of 27,000 messages per second, and over 95% closed-loop execution success rate. These results indicate that the platform meets its design targets in prediction accuracy, response latency, and scheduling efficiency under real-world conditions, providing a foundation for informatization and intelligent upgrading in urban rail transit.