Application of novel framework based on ensemble boosted regression trees and Gaussian process regression in modelling thermal performance of small-scale Organic Rankine Cycle (ORC) using hybrid nanofluid

This work examined the thermal performance of a small-scale solar organic Rankine cycle system, in which a flat plate solar collector was employed to supply heat to the organic Rankine cycle system. In the ORC system, WO₃+MWCNT/water nanofluid was employed for the solar collector, while MWCNT/R141b nano-refrigerant was used in the ORC system. In the first stage, the thermal and exergy efficiency of the ORC system was experimentally examined at different flow rates and concentrations of nanofluid. As a result, the peak thermal efficiency of 73.21% was attained at 3 L per minute (lpm) of nanofluid flow rate and 1.5 vol% of nanofluid concentration. Moreover, a significant increase in energy and exergy efficiency by 8.52% and 6.30%, respectively, was observed for 0.5 vol% of nanofluid concentration and 3 lpm of nanofluid flow rate in the collector. In the second stage, the experimental data was utilized to develop contemporary ensemble machine learning algorithms such as boosted regression trees (BRT) and Gaussian process regression (GPR) aiming to predict the solar collector's energy efficiency and the ORC system's energy-exergy efficiency. The GPR-based models performed exceptionally well as Pearson's coefficient (R), coefficient of determination (R²), mean squared error (MSE), and root mean squared error (RMSE) were 0.9999, 0.9998–0.9999, 0.0015–0.0524, and 0.0124–0.0723, respectively. For BRT-based model the R, R², MSE and RMSE were 0.9604–0.9949, 0.97–0.99, 0.0969–0.8783, and 0.3113–0.9372. Theil's U2 was employed to calculate the uncertainty in the prediction framework, which was determined to be in the range of 0.002–0.00369 for GPR and 0.00228–0.1414 for the BRT-based model. Overall, both machine learning algorithms performed well, although GPR did better than BRT across the board.