为精准把控并及时调节葡萄大棚棚内小气候，利用清徐县葡萄大棚农田小气候站观测数据及气象站、辐射站、土壤水分站资料，建立以棚外气温、相对湿度、风速、总辐射、土壤湿度为输入变量，棚内气温、相对湿度、土壤温度为输出变量的基于BP神经网络葡萄大棚小气候预测模型。为了对比分析BP神经网络的精确度和稳定性，同时建立多元线性回归模型。结果表明，基于BP神经网络建立的预测模型，其训练值和实测值之间的绝对误差分别为1.55 ℃、4.46%、0.77 ℃，标准误差分别为2.18 ℃、5.94%、1.00 ℃；预测值和实测值之间的绝对误差分别为1.37 ℃、2.84%、0.42 ℃，标准误差分别为1.96 ℃、4.60%、0.53 ℃。预测效果明显优于多元线性回归模型，预测精度满足棚内小气候要素预报要求。

To precisely control and adjust the microclimate inside grenhouses for grape production, with the data of farmland microclimate station, meteorological station, radiation station and soil moisture station, a simulation and forecast model of Back propagation neural network inside the grape greenhouse in Qingxu was built, which took the temperature, relative humidity, wind speed, total radiation and soil moisture content outside the greenhouse as input variables, and the temperature, relative humidity and soil temperature inside the greenhouse as output variables. In order to compare and analyse the accuracy and stability of Back propagation neural network, a multiple linear regression model was built at the same time. The results showed that the absolute errors of the model of Back-propagation neural network, when compared training values with the measured values, were 1.55 ℃, 4.46% and 0.77 ℃, respectively, and the RMSE values were 2.18 ℃, 5.94% and 1.00 ℃, respectively. The absolute errors of the model, when compared the predicted values with the measured values, were 1.37 ℃, 2.84% and 0.42 ℃, respectively, and the RMSE values were 1.96 ℃, 4.60% and 0.53 ℃, respectively. The prediction effect was obviously better than that of the multiple linear regression model, and the prediction accuracy of the model was of satisfactory for the microclimate prediction inside grenhouses.

S639；P463.4

山西省科学技术厅重点研发计划项目(201803D221001-3)。