参考作物蒸发蒸腾量(ET0)是计算作物需水量和进行灌溉预报的基本要素.本文利用天气预报可测因子和Penman Monteith(PM)公式ET0计算值作为基础数据,分别建立BP神经网络模型和ANFIS自适应模糊神经推理系统模型,两种模型的估算值与PM公式的计算值没有明显差异,均表现出显著的相关性以及整体吻合度.本文对两种模型取相同的数据样本进行比较,BP-ET0预测结果的MRE值为32.13%,RMSE为0.134 mm,而R2达到了0.971,说明模型预测精度高,稳定性良好.相较于ANFIS-ET0的检验结果,BP-ET0模型的均方根误差更小(0.134 mm/d<0.188 mm/d),表明其预测精度更高;而ANFIS-ET0模型估算值的平均相对误差明显小于BP-ET0模型估算值(16.92%<32.13%),显示出ANFIS-ET0模型更高的稳定性.两种预测模型的输入项完全可以从当前短期天气预报因子中取得而不需要专用测量设备,程序操作简单,具有实用价值,为实时灌溉预报提供了理论基础.%Reference crop evapotranspiration (ET0) is a basic property to calculate crop water requirement and irrigation scheduling. The weather forecast measurable variables and ET0 calculated by Penman Monteith (PM) formula were used as input to develop the BP neural network model and ANFIS adaptive neural fuzzy inference system model. The estimated value of the two models and the ET0 calculated by the PM formula was not significantly different, showing signiifcant correlation and overall degree of agreement. The two models were compared with the same data samples. The MRE value of BP-ET0 prediction was 32.13%, RMSE was 0.134 mm, andR2 was up to 0.971, which shows that the model has high prediction accuracy and good stability. Compared with the ANFIS-ET0 test results, the root mean square error of BP-ET0 model (0.134 mm/d) was smaller than that of ANFIS-ET0 model (0.188 mm/d) which means that the prediction accuracy was higher for the former model. The average relative error of the ANFIS-ET0 model (16.92%) was less than that of the BP-ET0 model (32.13%) which showed higher stability of the ANFIS-ET0 model. The input items of the two forecast models can be obtained from the current short-term weather forecast factors without special measuring equipment and the procedure is simple, thus it has high practical value and provides a theoretical basis for real-time irrigation forecasting.
展开▼
