Employing Machine Learning Regression Models for Performance Prediction of a Two-Stage Solar Dryer

The present manuscript aims to examine the application of artificial intelligence techniques to predict the performance of a two-stage solar dryer system equipped with preheating via two solar air collectors. Three machine learning algorithms, namely Linear Regression, k-Nearest Neighbors, and Multilayer Perceptron, were developed, trained, and evaluated on experimental data to identify the most accurate model for predicting the system's performance throughout the year. The dataset undergoes a preprocessing step including temporal linear augmentation, z-score normalization, and second-degree polynomial feature expansion. The evaluation process utilized several performance indices, including mean square error, mean absolute error, mean absolute percentage error, and the coefficient of determination. The k-nearest neighbors algorithm emerged as the most precise model for predicting moisture removal and the gain output ratio of the solar dryer based on ambient temperature and solar radiation across all performance indicators. This model was subsequently employed to forecast the system's performance across all months of the year, revealing a peak in average daily moisture removal of 9.40 kg/day in June and a minimum value of 4.47 kg/day in November, corresponding to increased solar radiation and daily exposure hours during the summer months compared to winter. Moreover, the average gain output ratio throughout the year was found to be approximately 0.42.