China's 1st runoff forecasting AI large model launched in Guizhou

The AI large model marks a breakthrough in integrating meteorological and runoff forecasting. [Photo provided to chinadaily.com.cn]

The "Wujiang Ruisuan" AI large model was launched in Southwest China's Guizhou province on April 29. Developed by Guizhou Wujiang Hydropower Development Co in collaboration with Huawei and Guodian Nanjing Automation Co, and built upon the Huawei Cloud Pangu Meteorological Model, this marks the first time AI large model technology has been applied to the field of runoff forecasting.

The model utilizes spatio-temporal graph neural networks to remodel runoff generation mechanisms, capturing the deep coupling between rainfall and dynamic hydrometeorological features.

Compared with traditional mechanistic models, short-term runoff forecasting accuracy has improved by more than 5 percentage points, while inference efficiency has reached the second-level, or real-time, scale.

Huawei's Hybrid Cloud Intelligent Computing Base has established a cloud-edge collaborative AI architecture in which AI development and training are managed at the group headquarters, while scenario-specific inference is deployed at Wujiang stations.

Through software-hardware collaborative scheduling and end-to-end security, the system masks underlying hardware complexity. This supports the rapid construction and optimization of the model across multiple scenarios, including meteorology, runoff, and power forecasting.

This platform-based approach ensures that these AI capabilities are not limited to the Wujiang River but can be replicated across regions and industries.

Based on a high-quality dataset exceeding 50 terabytes and covering three major categories and over 60 types, the system can support a wide range of fields, including flood control, drought relief, ecological protection, and transportation logistics.