Prediction of Drying Kinetics and Microclimate Conditions in a Tunnel-Type Solar Dryer Using Multilayer Perceptron Neural Networks

Abstract

This research established an experimental database characterizing the solar drying behavior of seven distinct agricultural products (including fruits, vegetables, and meat) in a tunnel-type solar dryer. Based on this experimental data, a dual-model architecture using multilayer perceptron (MLP) neural network was developed to predict both the internal microclimate and the drying kinetics. The first network (ANN 1) mapped meteorological variables to the dryer’s internal conditions, while the second (ANN 2) predicted moisture loss. The results demonstrate distinct predictive capabilities for each physical phenomenon: the thermodynamic model (ANN 1) captured stochastic weather fluctuations, with an R2 of 0.9878 and an MAPE of 4.64%. The kinetic model (ANN 2) achieved near-perfect linearity with an R2 of 0.9997 and an MAPE of 0.49%, significantly outperforming baseline linear regression models (R2 approx 0.78). These findings confirm the system’s capacity to generalize across diverse food types and variable weather conditions, providing a robust tool for future Model Predictive Control (MPC) applications.