A CNN-based temperature prediction approach for grain storage

Temperature prediction has a pivotal role in the grain storage phase. Accurate prediction results can optimise the effect of ventilation decisions and reduce the losses of stored grain. Most existing studies have only focused on layer temperature predictions whose predict particle size is very large. In contrast, this paper attempts to use convolutional neural network (CNN) to predict the point temperature of grain piles. The CNN-based approach uses multiple convolution kernels that share weights to capture the characteristics of grain temperature at different locations, which make full use of the temperature information around the target point. Experiments on real business data show that compared to other conventional algorithms, CNN has the best prediction effect on point temperature prediction problems. © 2020 Inderscience Enterprises Ltd.