Abstract: This study investigated the effects of the intrinsic properties of sorghum for Baijiu production on its hydration process in order to enable prediction of the moisture content of different varieties of sorghum during soaking. The physicochemical properties of 23 cultivars of Baijiu sorghum were measured, and the kinetic process of their hydration was analyzed upon soaking at a constant temperature of 40 ℃. Besides, the correlation between the hydration kinetics characteristics (initial hydration rate and equilibrium moisture content) and physicochemical properties of sorghum (seed coat thickness, hardness, specific surface area, protein, fat, tannin, starch, amylose, amylopectin) was analyzed. It was found that the hydration kinetics characteristics of sorghum were correlated with the specific surface area, hardness, fat, tannin, amylose and amylopectin. A backward propagation (BP) neural network model with 10 nodes in the hidden layer was established using the hardness, specific surface area, fat, tannin, amylose, soaking time and initial moisture content as the input layer, and the moisture content of sorghum as the output layer. Using the Levenberg-Marquardt (L-M) algorithm as the training function and tansig-purelin as the network transfer function, the BP neural network model was obtained after finite training. The correlation coefficient (r) between the predicted and the experimental values of the moisture content was 0.99, and the mean square error (MSE) was 0.02. This BP neural network model was capable of predicting the moisture content in different varieties of Baijiu sorghum during the soaking process. This research provides a theoretical foundation and technical support for the further development and precise control of the soaking process.

Key words: Baijiu sorghum; physicochemical properties; hydration kinetics; backward propagation neural network; water content prediction