A Performance Prediction Solution for Modified Hemispherical Solar Still Based on Machine Learning

Abstract

Artificial intelligence has been involved into different research fields. One of the interesting fields is the mechanical engineering field. This research work intends to introduce an alternate prediction approach that can predict the Hemispherical Solar Still (HSS) performance effectively without using an empirical method. The thermal performance of the HSS is predicted using five prediction models including Decision Tree (DT), Random Forest (RF), Gradient Boost (GB), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), which are performed, assessed, and compared. The proposed prediction models are built using real experimental data that has been recorded. The effective prediction model to be employed in the prediction of the hourly productivity and the instantaneous efficiency of the HSS is determined using five statistical error values. The comprehensive comparative analysis that was carried out demonstrates that, as compared to other models, the Decision Tree model may be used to estimate the thermal performance of solar stills without the need for additional experiments, saving money, effort, and time as its R2 and EVS values were near to one and the other statistical values (MSE, NAE, and Median Absolut Error) were very small. According to the results of the experiment, the HSS has an average hourly productivity of 0.478 L/m2 with the daily water production of 0.477, 0.465, 0.476, 0.477 L/m2 day for the proposed DT, SVM, RF, KNN and GB models, respectively. In addition, an average instantaneous efficiency of 45.199% with daily efficiency 45.188, 46.374, 46.375. 44.794 and 45.2 % for DT, SVM, RF, KNN and GB, respectively. Therefore, the proposed models appear a superior performance prior to performance estimation of HSS and can be considered as an efficient solution for this issue.