The moisture content of tobacco is a key quality indicator in tobacco leaf shredding and drying processing. However, due to the delayed measurement of moisture in production and the complex coupling relationship of production control parameters, existing moisture prediction methods often fail to meet expectations. In order to accurately predict the moisture content of tobacco in the tobacco shredding and drying workshop, we propose a tobacco moisture content prediction model called MoiConNet based on multi-scale temporal neural networks. This model utilizes Temporal Convolutional Network (TCN) and Gate Recurrent Unit (GRU) to extract features of signals at different time scales. After fusing these temporal features with non-temporal industrial monitoring parameters, they are fed into a Deep Neural Network (DNN) to predict tobacco moisture content. The experimental results on multiple tobacco process datasets of the cigarette factory’s tobacco shredding and drying production line show that the MoiConNet exhibits superior generalization performance in predicting moisture content tasks, and the prediction accuracy significantly exceeds traditional machine learning methods and deep-learning-based prediction algorithms.