Background
As more large-scale solar arrays (utility-scale) are connected to the energy grid, predicting solar irradiance (e.g., the amount of sunlight hitting the panels) becomes crucial for accurately forecasting the power of these photovoltaic (PV) arrays. This is because solar power generation is highly dependent on sunlight, which is variable and unpredictable.
In recent years, temporal convolutional networks (TCN), a type of deep learning model, have become a state-of-the-art method for sequence modeling problems. Similar to the spatial invariant property of conventional neural network architectures, TCNs have a time in variant property and can learn a temporal pattern from any time in a data series.
Invention Description
Researchers at Arizona State University have developed a cohesive system for predicting the amount of solar radiation hitting a horizontal surface of Earth by combining the most useful parts of multiple existing systems and methods. This system involves creating a new TCN architecture by combining 3 TCN modules. This allows for the accurate determination of the set of weather features mostly closely correlated to irradiance.
Potential Applications:
- Power utilities (e.g., consistent power grids)
Benefits and Advantages:
- Shorter run times
- Increased accuracy for predictive capability
- Improved performance