A coupled model of GM( 1,1 ) and BP neural network is established and applied in groundwater dynamic simulation and prediction of Minqin basin using data of well 64, well 65 and well 84 as representatives. Accuracy test of this model indicates that the average relative errors of these three wells are 0.45 ％, 0.93 ％ and 0.62 ％ respectively, which are all less than 1％ and meet the accuracy requirements. Compared with GM( 1,1 ), the predicted average relative errors of this coupled model are all less than that of GM( 1, 1 ) model. And when compared with BP model, the average absolute error of predicted groundwater depth of 1998 ～ 2001 reduced greafly from 0.32 m to 0.07 m. The results show that this coupled model considers the certainty and uncertainty of groundwater depth time series, with high accuracy of possessing prediction, and can be used for short term prediction.%先建立等维新患GM(1,1)模型和BP神经网络模型相耦合的灰色BP神经网络组合模型,再以民勤盆地64、65和84号井为代表,运用此模型模拟和预报石羊河下游民勤盆地的地下水埋深动态.模型精度检验表明,64、65和84号井预测值的平均相对误差分别为0.45%,0.93%,0.62%,均小于1%,符合精度要求.相比GM(1,1)模型,组合模型预测的相对误差整体上较小;相比BP模型,64号井组合模型预测的1998～2001年地下水埋深平均绝对误差从0.32m减少为0.07m,精度显著提高.结果表明:组合模型综合考虑了地下水埋深序列的确定性和不确定性变化,具有更高的预测精度,适合于短期预报.
展开▼
