The sun is becoming an increasingly important source of electricity. Exact sunlight forecasts developed by A * STAR researchers could significantly improve the performance of solar power plants, making it a viable alternative to coal-based energy sources.
A photovoltaic power plant can cover up to 50 square kilometers of Earth’s surface and produce up to one billion watts of electricity. This capacity depends on the amount of sunlight in this location, so the ability to predict solar radiation is vital to find out how much power the plant will contribute to the network on a particular day.
“Forecasting is a key step in the integration of renewable energy into the grid,” says Dazhi Yang from A * STAR Singapore’s Institute of Manufacturing Technology (SIMTech). “It is an emerging subject that requires a wide range of interdisciplinary knowledge, such as statistics, data science and mechanical learning.”
Yang, along with Hao Quan from the Experimental Power Grid Center and colleagues from Tennessee University and the National University of Singapore, have developed a numerical approach to weather forecasting that effectively combines multiple sets of data for improved predictions associated with the accuracy of solar radiation.
Hourly changes in the atmosphere, annual changes in the distance between the Earth and the Sun or 10-year changes in the Sun’s inner circles can change the amount of sunlight reaching the surface of the Earth. These changes occur at very different time scales, so conventional prediction methods bring volatility, making data processing via computers easier. However, these methods are based on a simple prediction process that does not have a very heavy weight, so the forecasts are not accurate. In addition, the forecasts produced by the conventional methodology are accurate only on the timetable.
Yang and the team developed a framework that reconciles with different time scales, forming a hierarchical time frame that gathers predictions received in different timeframes, such as high frequency data, hourly data, and low frequency daily data. “Temporal reconciliation is a type of prediction model of the set that accurately predicts daylight-based solar illumination, using data of different timing details, hourly, of two hours and on a daily basis,” Yang explains. “These different forecasts are then perfectly matched through statistical models to produce a final forecast”.
Researchers looked at the numerical weather forecasting method using data from 318 photovoltaic plant locations in California for one year. The Temporal reconciliation method proved to have significantly outperformed the other numerical predictions for predicting sunlight the next day.