The Sun is turning into an undeniably significant source of clean electricity. A*STAR scientists are developing accurate sunlight forecasts to significantly enhance the performance of solar energy plants, making it a reasonable option to carbon-based sources of power.
Up to 50 square kilometers of the Earth can be covered by a photovoltaic power plant which can generate up to a billion watts of electricity. However, this capacity highly depends on whether there is enough sunlight at the location. Hence, being able to predict solar irradiance is significant for knowing how much power will be contributed by the grid daily.
Dazhi Yang who is from the A*STAR's Singapore Institute of Manufacturing Technology (SIMTech) says, “Forecasting is a key step in integrating renewable energy into the electricity grid.” He also added that it is a rising subject that requires a wide range of cross-disciplinary information, for instance; insights, machine learning or information science.
Dazhi Yang along with his colleagues from the University of Tennessee at Chattanooga and the National University of Singapore, and Hao Quan, from the A*STAR Experimental Power Grid Centre, have taken to develop a numerical approach to weather analysis and prediction. This approach has led to efficiently combining numerous datasets to enhance the exactness of solar irradiation forecasts.
The changes in the sun’s internal cycles every ten years alter the amount of sunlight that reaches the surface. Since these changes happen at different time scales, conventional forecasting methods model variability separately at distinctive timescales. Therefore, this approach makes computer processing much more straightforward and more accessible. Nonetheless, these techniques depend on a direct addition of forecasts, with no weighting that improves more utilization of better forecast sub-series.
Yang's team developed a framework that reconciles different timescales. They achieved this by creating a temporal hierarchy that acquires forecasts obtained at various timescales such as daily data, hourly data, and low and high-frequency data. These distinctive forecasts are optimally combined through statistical models, producing a final prediction.
The researchers have tested their numerical weather prediction method utilizing data from 318 photovoltaic power plant sites situated in California. Their temporary compromising strategy significantly outperformed various other numerical day-ahead forecasts.