China Launches First Large-scale Watershed Runoff Prediction AI Model for Smart Hydropower Scheduling

Xinhua News Agency reports that China’s first large-scale intelligent model dedicated to watershed runoff prediction, known as Huadian Zhiyu Wujiang Ruisuan, has officially gone online with an integrated smart scheduling platform for hydropower, wind and solar energy. The launch marks a new phase of intelligent operation and refined scheduling for China’s hydropower watershed management.

Developed under China Huadian’s intelligent AI system framework, the innovative model integrates advanced artificial intelligence technologies with professional hydropower operational scenarios. It effectively addresses long-standing industrial challenges, including inaccurate medium and long-term runoff forecasting and inefficient coordinated scheduling of complementary water, wind and solar power resources. As a pioneering application of large-model technology in hydrological prediction across complex karst landform watersheds, the system fills key technical gaps in the industry’s intelligent forecasting capabilities for sophisticated terrains.

The intelligent platform delivers notable improvements in forecasting precision. It boosts the accuracy of short-term flood prediction and 15-day medium-and-long-term runoff forecasting by 5 percent compared with traditional conventional models. The upgraded prediction performance provides solid intelligent support for flood prevention and disaster mitigation, ecological environment protection, waterway transportation and logistics, and other water resource-related fields. It drives a fundamental shift in watershed management from experience-based judgement to data-driven and intelligent decision-making.

The project is built on an industry-university-research collaborative ecosystem jointly established by China Huadian and Huawei. Leveraging Huawei Cloud and the Pangu Meteorological Large Model, the team has constructed a multi-source fusion dataset exceeding 50 terabytes. The platform adopts a suite of core technological innovations, including multi-modal data fusion, spatio-temporal inference algorithms, deep learning-based rainfall-runoff coupling modelling, and flood prediction methods optimised with recurrent neural network parameter identification. It also introduces large language model technologies into hydrological forecasting scenarios.

These integrated technical upgrades enable the system to achieve a medium-term runoff prediction accuracy rate of over 85 percent for 15-day forecasts. The intelligent scheduling system maximises the consumption efficiency of clean energy and optimises overall power generation benefits across the watershed. Certified as one of the first strategic high-value artificial intelligence scenarios for central state-owned enterprises, the new platform sets a solid technological benchmark for the digital transformation of China’s hydropower industry.

Moving forward, the intelligent watershed management solution will continue to empower the high-quality development of the new energy system. It supports safer, greener and more efficient utilisation of water resources and clean energy, promoting the iterative upgrading of intelligent scheduling technologies for multi-energy complementary watersheds nationwide.