Accurate and Scalable Receiver-Level Flux Prediction: A Fully Data-Driven Solution

Abstract

Concentrated Solar Technologies (CST) systems, particularly central tower configurations with heliostat fields, play a critical role in the renewable energy landscape. By focusing sunlight from thousands of heliostats onto a central receiver, these systems generate high-temperature heat, which serves as a key resource for dispatchable power generation and industrial processes. Accurate receiver-level flux prediction, which depends on precise heliostat characterization, is essential for optimizing efficiency and operational control. However, existing characterization methods face trade-offs between accuracy and scalability, limiting their practicality for large-scale deployment. To overcome these limitations, this study introduces a fully data-driven framework that unifies heliostat characterization and flux prediction, leveraging operational data from standard calibration procedures. Expanding upon previous work that employed StyleGAN for beam-characterization-based predictions, this approach advances the methodology to achieve accurate receiver-level flux predictions. While the prior method demonstrated a proof of concept for a unified data-driven approach, it remained constrained to flux predictions on the calibration target itself. This study introduces key innovations, including aim point generalization strategies and a novel receiver projection technique, effectively bridging the gap between beam-characterization-based heliostat characterization and accurate receiver-level flux predictions. The proposed Transformer-based architecture achieves receiver-level focal spot prediction errors below 12%, exceeding the accuracy of state-of-the-art deflectometry-enhanced ray tracing. By relying exclusively on standard calibration images, the method remains both cost-efficient and scalable, offering a practical solution for large-scale CST applications.