Research on Storage Temperature Prediction of Imported Corn Based on the ARIMA-LSTM Algorithm
摘要
针对进口玉米储藏过程中温度变化非线性特征显著及传统预测模型的局限性,提出了一种结合自回归移动平均模型(Autoregressive Integrated Moving Average Model,ARIMA)与长短期记忆神经网络(Long Short-Term Memory,LSTM)的粮温预测组合模型.首先,利用 ARIMA 模型对玉米储藏温度的时间序列数据进行线性特征提取,捕捉数据的趋势和季节性规律.结合预测结果,进一步组合大气温度、湿度、仓温及仓湿等多源环境参数作为输入特征,构建 LSTM 深度神经网络模型.利用 LSTM 网络强大的非线性拟合能力,捕捉残差中的复杂模式,将线性预测与非线性校正相结合,显著提升模型预测精度.试验结果表明,相较于单一的 ARIMA 模型,ARIMA-LSTM 组合模型预测精度显著提升.39 号仓和 43 号仓第1~4 层的平均绝对误差(Mean Absolute Error,MAE)分别降低了 0.14、0.09、0.09、0.09 和 0.11、0.06、0.05、0.05;均方根误差(Root Mean Square Error,RMSE)分别降低了 0.17、0.09、0.09、0.10和0.13、0.06、0.04、0.06,验证了该模型在复杂储藏环境下的适用性.本研究可为粮食储藏温度管理提供新的技术手段,对保障粮食储藏安全、减少霉变损耗具有重要应用价值.
Aiming at the complex nonlinear characteristics of temperature changes during the storage of imported corn and the limitations of traditional prediction models,a hybrid model of grain temperature prediction combining the Autoregressive Integrated Moving Average(ARIMA)model and the Long Short-Term Memory(LSTM)neural network was proposed.Firstly,the ARIMA model was used to extract the linear features of the time series data of corn storage temperature,and the trend and seasonal patterns of the data were captured.Then,based on the pre-diction results of the ARIMA model,the multi-source environmental parameters such as atmospheric temperature,humidity,warehouse temperature and warehouse humidity were further combined as input features to construct the LSTM deep neural network model.Using the powerful nonlinear fitting ability of the LSTM network,the complex patterns in the residuals were further captured,and the linear prediction is combined with the nonlinear correction.The model significantly improved the prediction accuracy.The experimental results showed that in the prediction of corn storage temperature,compared with the single ARIMA model,the ARIMA-LSTM hybrid model significantly improved prediction accuracy.the mean absolute error(MAE)of the first,second,third and fourth layers of No.39 warehouse and No.43 warehouse was reduced by 0.14,0.09,0.09,0.09 and 0.11,0.06,0.05,0.05 respectively.The root mean square error(RMSE)was reduced by 0.17,0.09,0.09,0.10 and 0.13,0.06,0.04,0.06,respectively,which verified the applicability of the model in complex storage environments.This study provides a new technical means for grain storage temperature management,which is of great practical signifi-cance for ensuring safe grain storage and reducing losses caused by mold spoilage.