地下水文预测中BP网络的模型结构及算法探讨

本文探讨了人工神经网络中不同BP网络结构和算法在区域地下水文预测中的应用，实例比较了不同层次结构、学习速率、隐层单元数及不同算法等对收敛效果、模拟预报结果的影响。提出了一些BP模型的设计应用技术，即学习速率的取值范围与BP网络层数有一定关系，层数多，稳定区间较小，一般学习速率取值为0.01～0.1。快速BP算法从训练速度，收敛精度等方面均优于普通BP算法，可作为改进BP算法之一。在此基础上根据黄河河套灌区多年的水文、气象和地下水信息，对灌区多年的年地下水埋深变化进行了模拟，预测了河套灌区节水工程实施后未来灌区地下水位下降的趋势，为大型灌区节水工程改造与BP模型在区域地下水文中的应用提供了参考。

Structure and algorithm of BP Network for underground hydrology forecasting

The BP networks with different structure and algorithm are applied to forecast regional groundwater level.The layers of structure,learning rate,number of hidden unit and BP algorithm which affecting the convergence and the accuracy of simulation are studied.It is found that the more the layer unmber,the more narrow the stable interval.Normally,the learning rate is ranging from 0.01 to 0.1.The accelerated BP algorithm is better than normal BP both in training speed and convergence accuracy.Two kinds of BP algorithm are used to simulate the annual groundwater level variation of Hotao district in Yellow River basin,based on long term hydrological and climate data.The prediction of groundwater in this district,in case the water saving project is implemented,is carried out.