Developing deep learning-based model for silicon-based solar cells in concentrator photovoltaic systems: A real-time prediction for efficient application-oriented performance

Concentrator photovoltaic (CPV) technology harnesses intense incident solar radiation, offering the potential for simultaneous electrical power generation and thermal utilization via compact, cost-effective heat sinks. However, maximizing the concentration ratio necessitates intensive cooling, resulting in low-grade heat generation. On the other hand, to achieve the demanded temperature of this low-grade heat generation for thermally driven applications, several operational and design parameters, including concentration ratio and heat sink characteristics, need to be harmonized. This can be numerically revealed using the conventional finite volume method (FVM) through optimization techniques/intensive parametric studies for wide-range concentration ratios under different cooling techniques which needs a prohibited computational cost and time. Addressing this challenge, the present work develops a deep learning-based model as a computationally efficient alternative for real-time performance prediction of silicon-based solar cells. The model is trained and validated using extensive datasets from a numerically and experimentally validated 3D thermal-fluid FVM model. These datasets cover wide variations in concentration ratios, heatsink heat transfer coefficients, meteorological conditions (ambient temperature and wind speed), cell reference characteristics (reference efficiency and temperature coefficient), and cell structure providing a comprehensive input-output mapping. The optimized neural network demonstrates high accuracy and reliability with a minimal mean square error and a coefficient of determination approaching unity. Furthermore, a user-friendly software with a graphical user interface (GUI) is developed, enabling two modes of analysis: real-time performance optimization through dynamic design parameter adjustments and real-time solutions for massive parametric studies. This novel workflow significantly reduces computational costs and processing times, facilitating instantaneous generation of characteristic performance maps (CPMAPs). The proposed approach accelerates decision-making for CPV applications and can be extended to other energy-related technologies, offering a transformative tool for both industry and research communities.