Predicting equilibrium moisture content of mushrooms by NARX neural network and first order kinetic modelling approaches

Abstract

Soft-cellular and porous agricultural products continuously exchange water vapor with their surrounding environment. The aim of this work was to model the water vapor sorption kinetics of mushrooms by strictly pragmatic and numerical approaches. A nonlinear autoregressive with exogenous inputs (NARX) neural network was used to simulate water vapor sorption kinetics over a wide range of relative humidity at temperatures of 20-40°C. The predictive power of the neural network was based on physical data, namely time-series of relative humidity at constant temperature. In addition, the time dependence of water vapor sorption was fitted by a first-order kinetic model, which was applicable for all relative humidity steps. Both models were fed with a total sample size of ca. 11000 data points generated by an automated gravimetric dynamic vapor sorption system. A neural network with a seven-node hidden layer was found to have the best performance. The results revealed the potential efficient use of artificial neural networks in simulating water vapor sorption kinetics of mushrooms and consequently predicting equilibrium moisture content with a high accuracy in the entire range of relative humidity and temperature covered by the experiments.