A Novel Prediction-Optimization Machine Learning Framework for Nanofluid-Based Photovoltaic/Thermal Systems.

Area of Science:

• Solar Energy Engineering
• Materials Science
• Computational Science

Background:

• Nanofluid-based spectral filtering enhances photovoltaic/thermal (PV/T) systems by utilizing the full solar spectrum.
• Optimizing PV/T systems with nanofluids is complex due to nonlinear parameter interactions.

Purpose of the Study:

• To develop an integrated prediction-optimization framework for designing high-performance nanofluid-based PV/T systems.
• To accurately analyze multi-parameter interactions and achieve globally optimal designs.

Main Methods:

• Constructed high-throughput datasets using theoretical calculations (Lorentz-Mie theory, Monte Carlo simulations, coupled opto-electro-thermal model) for Ag, Au, and Al nanofluids.
• Employed deep neural networks (DNN), random forest (RF), and decision tree (DT) models for performance prediction.
• Integrated machine learning models with genetic algorithms (GA) for a closed-loop prediction-optimization process.

Main Results:

• The DNN model achieved prediction accuracies above 99.48% for key PV/T performance indicators (ηpv, ηth, MF), outperforming RF and DT models.
• SHAP analysis quantified feature contributions, enhancing model interpretability.
• The DNN-GA framework identified globally optimal design parameters for Ag, Au, and Al nanofluids, demonstrating practical value.

Conclusions:

• The developed DNN-GA framework offers an innovative and efficient approach to designing nanofluid filters for PV/T systems.
• This method reduces reliance on iterative experimentation, accelerating the development of advanced solar energy technologies.