Adversarial ML super-resolves climate wind, solar data

July 10, 2020 //By Rich Pell
Adversarial ML super-resolves climate wind, solar data
Researchers at the U.S. Department of Energy’s (DOE’s) National Renewable Energy Laboratory (NREL) say they have developed a novel machine learning (ML) approach that can quickly produce up to a 50x resolution enhancement of climatological wind and solar data.

Based on adversarial machine learning , the approach, say the researchers, can quickly enhance the resolution of wind velocity data by 50 times and solar irradiance data by 25 times - an enhancement that has never been achieved before with climate data - and will enable scientists to complete renewable energy studies in future climate scenarios faster and with more accuracy.

"To be able to enhance the spatial and temporal resolution of climate forecasts hugely impacts not only energy planning, but agriculture, transportation, and so much more," says Ryan King, a senior computational scientist at NREL who specializes in physics-informed deep learning.

Accurate, high-resolution climate forecasts are important for predicting variations in wind, clouds, rain, and sea currents that fuel renewable energies. Short-term forecasts drive operational decision-making; medium-term weather forecasts guide scheduling and resource allocations; and long-term climate forecasts inform infrastructure planning and policymaking.

However, say the researchers, it is very difficult to preserve temporal and spatial quality in climate forecasts, and the lack of high-resolution data for different scenarios has been a major challenge in energy resilience planning. While various machine learning techniques have emerged to enhance the coarse data through super resolution - the classic imaging process of sharpening a fuzzy image by adding pixels - until now, no one had used adversarial training, which is a way of improving the performance of neural networks by having them compete with one another to generate new, more realistic data, to super-resolve climate data.

The researchers used the approach to improve the physical and perceptual performance of their networks. With the approach, the model produces physically realistic details by observing entire fields at a time - providing high-resolution climate data at a much faster rate.

The researchers trained two types of neural networks in the model - one to recognize physical characteristics of high-resolution solar irradiance and wind velocity data and another to insert those characteristics into the coarse data.

Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.