Multi-criteria optimization of nanofluid-based solar collector for enhanced performance: An explainable machine learning-driven approach

This study presents a novel hybrid framework that leverages machine learning to enhance the performance of nanofluid-based solar collectors (NBSCs). The framework is designed to identify the optimal control variables required to meet multiple performance criteria (such as simultaneously maximizing outlet temperature, thermal efficiency, and optical efficiency). This study introduces an end-to-end multi-criteria optimization framework that combines numerical simulations with a Gaussian process regression (GPR) and genetic algorithm (GA) for designing optimized NBSCs. In this approach, a minimal number of random samples are selected using Monte-Carlo sampling to perform numerical simulations. The control variables of the system are varied within practical ranges, and key performance metrics such as outlet temperature [ T o (°C)], thermal efficiency ( η t ), and optical efficiency ( η o ) are recorded. The input and output data are utilized to develop a computationally efficient GPR model. The generalization capability of the developed explainable machine learning (xML) models allowed for various data-intensive analyses, including sensitivity analysis, uncertainty quantification, interactive influence of control variables, and multi-objective optimization. The proposed computational framework helped explore previously unknown territory, leading to the identification of optimal settings for simultaneously maximizing all the responses. The optimal parameters led to a simultaneous improvement in the responses, with a 23.44 °C rise in outlet temperature, a 37.48 % increase in thermal efficiency, and a 28.62 % boost in optical efficiency, compared to the base dataset. The developed framework is rigorously tested to ensure its robust generalization and its applicability to calibrate other physical systems. The results of this study offer valuable insights for designing optimal NBSCs with improved operational performance. • A novel computational framework is proposed for accelerated optimization of the NBSC. • The GPR model outperforms other ML models, with an R 2 of 0.99 and prediction errors within ±5 %. • MCS, FDM and GPR are integrated for data-intensive downstream tasks that would have otherwise remained unexplored. • A multicriteria attainment framework effectively proposed input feature settings for achieving multi-criteria goals. • The proposed framework enabled exploration of new territory through the implementation of an active learning mechanism.