Exploring the specific heat capacity of water-based hybrid nanofluids for solar energy applications: A comparative evaluation of modern ensemble machine learning techniques

Abstract

The current study aims to give insight into the usefulness of three underutilized yet exceptionally efficient machine learning approaches in estimating the specific heat capacity (SHC) of nanofluids: Gaussian process regression, extreme gradient boosting, and support vector machine. The study made use of three uncommon metal oxides-multiwall carbon nanotubes-water nanofluids. Several factors influencing the SHC of nanofluids were considered during model development, including volume concentrations (0.25 %–1.5 %), range of temperature (25–50 °C), the base fluid's SHC, the specific heat capacity of the nanofluid, the density of the base fluid, the density of the nanofluid, and the average diameter of the used nanoparticles. All three models predicted the SHC of hybrid nanofluids with high precision, suggesting that they are a viable alternative to the time-consuming, expensive, and complex experimental approaches for measuring the SHC of nanofluids. The extreme gradient boosting (R = 0.9975, mean absolute percentage error = 0.01 %) model outperformed the Gaussian process regression (R = 0.9973, mean absolute percentage error = 0.015 %) and support vector machines (R = 0.9952, mean absolute percentage error = 0.021 %) in terms of prediction performance. The Nash-Sutcliffe efficiency for Gaussian process regression, extreme gradient boosting, and support vector machines models were 0.9956, 0.997, and 0.992, respectively, while Theil's U2 was 0.0752, 0.072, and 0.1015, indicating that the three machine learning algorithms performed well in the precise estimation of SHC of test nanofluids. These prognostic models can help to reduce the time and resources needed for experimental trials.