Fast ground irradiance computations for agrivoltaics via physics-informed deep learning models.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Photovoltaic Technology

Background:

• Agrivoltaic systems integrate solar panels and agriculture, requiring precise sunlight distribution analysis.
• Accurate irradiation quantification is crucial for optimizing bifacial solar module performance and yield.
• Traditional raytracing simulations for complex agrivoltaic scenes are computationally intensive.

Purpose of the Study:

• To develop a computationally efficient surrogate model for calculating ground-level irradiation in agrivoltaic systems.
• To enable real-time irradiance calculations for photovoltaic (PV) tracker algorithms.
• To optimize agrivoltaic system design and operation for maximum crop and electrical yield.

Main Methods:

• A deep learning-based surrogate model was trained using physics-informed data from raytracing simulations.
• The model utilizes direct normal irradiance, diffuse horizontal irradiance, solar position, and system geometry as inputs.
• A generative regression approach with 3D scene encoding was employed for irradiance mapping.

Main Results:

• The surrogate model computes ground irradiance maps in 3 milliseconds, a four-order-of-magnitude speed improvement over standard raytracing.
• Achieved rapid, accurate calculation of sunlight distribution on ground and module levels.
• Demonstrated feasibility for on-the-fly raytracing in edge computing applications.

Conclusions:

• Deep learning surrogate models offer a significant speedup for irradiation calculations in agrivoltaic systems.
• This approach facilitates efficient management and optimization of PV systems, particularly for edge computing applications.
• Enables faster development and deployment of advanced photovoltaic tracker algorithms.