基于MODFLOW与ANN耦合的泉流量模拟研究

针对人类活动影响下岩溶地区泉流量难以预测的问题,基于地下水数值计算模型———MODFLOW和人工神经网络两者的优点,尝试将两者结合建立松散型耦合模型。以河南省安阳市小南海泉域的泉流量预测为例,探索耦合模型的原理和算法,并与单纯MODFLOW模拟的结果相比较。由确定性系数、相对误差和相关系数3个指标来看,MODFLOW模拟结果分别为0.79、4.98%和0.84,MODFLOW-ANN耦合模型的模拟结果分别是0.88、-1.22%和0.89。研究结果表明,耦合模型吸取了MODFLOW的地下水数值分析功能和人工神经网络的非线性逼近能力,能很好地模拟出泉流量峰和谷的变化,提高预报精度,可以用于模拟泉流量的动态变化过程,该研究对泉域岩溶地下水的进一步开发利用具有一定的参考价值和指导意义。

Integration of artificial neural networks with a numerical groundwater model for simulating spring discharge

The simulation of spring discharge is difficult in Karst areas under human activity.An integrated model(MODFLOW-ANN) was developed by combining the merits of the numerical groundwater flow model-MODFLOW and the Artificial Neural Network model(ANN).Based on the application of this model in the Xiaonanhai spring catchment of a karst region,Henan Province,the principles and algorithms of the integrated model were discussed,and the results were compared between MODFLOW and MODFLOW-ANN.The coefficient of determination,relative error and correlation were 0.79,4.98% and 0.84,respectively,for MODFLOW,and 0.88,1.22% and 0.89,respectively,for MODFLOW-ANN.The results showed that this integrated approach could take the advantage of the groundwater numerical analyzing capacity of MODFLOW and the nonlinear approximation ability of ANN,thus precisely predict the peaks and troughs of spring discharge.This model improved the predicting accuracy and was successfully applied to modeling the spring discharge dynamic.