AlibabaDharma Institute(Lakeside Labs) held a Decision Intelligence product launch in Beijing, officially releasing the Eight ViewsLarge weather model.
According to the official introduction, the model introduces regional multi-source data on the basis of the global meteorological model, and the temporal and spatial accuracy can reach up to 1 km * 1 km * 1 hour. By significantly improving the prediction performance of temperature, irradiation, wind speed and other key meteorological indicators, the eight-view meteorological model is the first to land on the new power system with a high proportion of new energy, helping the State Grid Shandong Power Control Center to successfully predict a number of extreme weather conditions, and the accuracy of new energy power generation and power load prediction has been increased to more than 96% and 98%, respectively.
Based on years of technical accumulation in the fields of mathematical modeling, time series prediction, and interpretable AI, the Decision Intelligence Lab of Dharmo Academy builds a regional high-precision weather forecast model using the self-developed global meteorological model as a base model. By integrating local station data, meteorological facts, radar images, satellite images, open-source terrain and other multi-source and multi-modal data, it enhances the granularity and accuracy of the forecast results, and realizes hour-by-hour weather forecast updates on a 1-kilometer grid.
The "global-regional" synergistic prediction weather model is formally named "Eight Observations", which means "Eight Insights, Observation of Ten Thousand Elements". According to the introduction, through the pre-training and twin MAE masked self-coder structure, the Eight Observations weather model provides better initialization parameters and learns robust feature representations hidden under highly fluctuating weather data, so as to achieve accurate grasp of the weather.
The actual data show thatThe prediction accuracies of the model in terms of regional irradiance, wind speed, cloudiness and temperature are improved by 40%, 27%, 24% and 11.8%, respectively, compared with the traditional weather forecasts.
On the basis of providing general modeling capabilities, the Octaview Meteorological Big Model will continue to improve its performance for key indicators such as cloudiness and precipitation, and is expected to provide a decision-making basis for more scenarios such as aviation early warning, agricultural production, and preparations for sports events.