基于GWO-BP神经网络及粮食压缩实验对粮食孔隙率的预测

孔隙率是影响粮堆内热湿传递的关键参数，为探究粮仓中散装粮堆孔隙率的分布规律，通过开展粮食压缩实验来获取不同的粮食种类在不同水分含量和竖向压力条件下的孔隙率。提出了一种基于灰狼算法优化BP(GWO-BP)神经网络的粮食单元体孔隙率预测模型，并将该模型与BP神经网络模型、随机森林模型的孔隙率预测结果进行对比，最后利用粮食单元箱实验对该模型的泛化能力进行验证。结果表明，GWO-BP神经网络模型的孔隙率预测性能最佳，该模型的评价指标R2为0.960 5、RMSE为0.013 7及MAE为0.0131，均在允许的范围内。本研究为粮食孔隙率的确定提供了一种神经网络预测的方法，为深入开展粮堆多场耦合分析提供了重要基础，为安全储粮提供了理论支持。

Prediction of Grain Porosity Based on GWO-BP Neural Network and Grain Compression Experiment

Porosity is a key parameter that affects the heat and moisture transfer within a grain pile. In order to investigate the distribution law of porosity in bulk grain piles in grain silos, grain compression experiment was carried out to obtain the porosity of different grain types under different moisture content and vertical pressure conditions. A porosity prediction model for grain cell based on GWO-BP neural network was proposed, and the model was compared with the porosity prediction results of BP neural network model and random forest model. Finally, the generalization ability of the model was verified using the grain cell box test. The results showed that the porosity prediction performance of the GWO-BP neural network model was the best, and the evaluation indexes of the model, including R2 of 0.960 5, RMSE of 0.013 7 and MAE of 0.013 1, were all within the permissible range. This study has provided a neural network prediction method for the determination of grain porosity, which could provide an important foundation for in-depth multi-field coupling analysis of grain piles and theoretical support for safe grain storage.