基于遗传算法的BP神经网络模型在地下水埋深预测中的应用——以蒙城县为例

地下水系统是一个高度复杂系统,针对地下水位与其影响因素之间的非线性映射关系,建立遗传算法优化BP神经网络浅层地下水埋深模型,对地下水埋深进行模拟和预测。使用RMSE、MAPE和NSE三种评价指标,将所得结果与BP神经网络和逐步回归模型进行对比。以蒙城县1974—1999年前期降雨量、前期地下水埋深和利辛县前期地下水埋深作为输入层,以当月地下水埋深作为输出层,将蒙城县2000—2010年地下水埋深作为检验样本,计算结果表明:遗传算法优化BP神经网络模型训练阶段和测试阶段RMSE分别为0.22和0.34、MAPE分别为7.6%和9.21%、NSE分别为0.89和0.85,泛化性能良好,有效规避了过拟合现象,且拟合和预测的精度较高。该模型可为地下水研究提供了一种有效浅层地下水埋深的预测方法,具有较好的应用前景。

Application of genetic algorithm-based BP neural network model to prediction of groundwater groundwater buried depth—a case study of Mengcheng County

The groundwater system was a highly complex system. A genetic algorithm-based optimal BP neural network model for shallow groundwater buried depth is established for the non-linear mapping relation between the groundwater level and its impacting factors，with which the groundwater buried depth is simulated and predicted. The results are compared with those from the BP neural network model and stepwise regression model with three evaluating indexes，i. e. ＲMSE，MAPE and NSE. By taking the antecedentrainfall and antecedent groundwater buried depth of Mengcheng County from 1974 to 1999 and the antecedent groundwater buried depth of Lixin as the input layers，taking the groundwater buried depth of the same month as the output layer and taking groundwater depth of Mengcheng County from 2000 to 2010 as the testing sample，the result shows that ＲMSE during the training phase and testing phase of the genetic algorithm-based optimal BP neural network model are 0. 22 and 0. 34 respectively，while MAPE are 7. 6% and 9. 21% and NSE are 0. 89 and 0. 85 respectively with better generalization performance，and then the over-fitting phenomenon is effectively avoided，while the relevant fitting and predicting accuracies are higher as well. This model can provide an effective method of predicting shallow groundwater buried depth for the study made on groundwater， which has a better application prospect.