Abstract: Aiming at the problems of unsatisfactory single-point prediction of grain temperature in a complex storage environment, and difficulty in meeting the needs of engineering applications in current temperature field modeling, based on the temperature field theory, and combined with the distributed temperature measurement system structure, this paper proposed a temperature field prediction model based on grain pile temperature data. The model was on the basis of BP neural network, used discrete temperature measurement points in the granary to predict the future temperature data of the corresponding points, and then used Kriging interpolation method for spatial interpolation, and used the temperature  of the known location to estimate the temperature  of the unknown point, and then establish predictive model of temperature field. The simulation  results showed that the average absolute percentage error of temperature prediction was 1.253 5%, and the root mean square error was 0.106 0. The prediction effect was good. When using Kriging interpolation method to interpolate temperature points, the average absolute percentage error was 9.470 0%, and the root mean square error was 0.865 1. Compared with the traditional single-point prediction algorithm for grain pile temperature, this model could better reflect the changing trend and temperature distribution of the temperature field in the granary, provide better data support for the granary manager, and realize auxiliary decision-making. The model was highly scalable and could be applied to various storage sites.

Key words: grain temperature, BP neural network, kriging interpolation, temperature field prediction

摘要： 针对复杂仓储环境中粮情温度单点预测效果不理想、现有温度场建模难以满足工程应用需求等问题，基于温度场理论，结合分布式测温系统结构，提出了基于粮堆温度数据的温度场预测模型。该模型基于BP神经网络，利用粮仓内离散测温点数据预测对应点的未来温度数据；再采用Kriging插值法进行空间插值，利用已知位置的温度值估计出未知点的温度值，进而建立温度场的预测模型。仿真测试结果表明，温度预测的平均绝对百分误差为1.253 5%，均方根误差为0.106 0，预测效果良好。采用Kriging插值法进行温度点的插值，其平均绝对百分误差为9.470 0%，均方根误差为0.865 1。对比于传统的粮堆温度单点预测算法，该模型能够更好地反映粮仓内温度场变化趋势以及温度分布的情况，为粮仓管理者提供更好的数据支持，实现辅助决策。该模型可扩展性强，能够适用于各种仓储现场。

关键词: 粮食温度, BP神经网络, Kriging插值法, 温度场预测