基于遗传算法的BP神经网络模型在地下水动态预测中的应用研究

地下水系统是一个复杂的非线性动力系统,地下水系统的输出(地下水位)与输入(降水入渗、蒸发、人工开采等)具有非常复杂的非线性关系。神经网络以其强大的处理非线性系统的能力而在地下水水位预测中得到了广泛的应用,它与传统的统计分析模型相比,具有更好的持久性和适时预报性,且能用于解决同时存在多个自变量和多个因变量的地下水系统预报问题。但是由于网络输入存在多重共线性,导致网络泛化能力不高,降低了网络的预测性能;并且在求解时易陷入局部极小,且收敛速度慢。针对以上问题提出了基于遗传算法的BP神经网络地下水动态预测模型。先用遗传算法优化确定BP网络的初始权阀值,然后应用LMBP算法在这个解空间里对网络进行精调,搜索出最优解或近似最优解。这样既实现了两者的优势互补,发挥了神经网络广泛的映射能力和遗传算法的全局搜索能力,也加快了网络的学习速度,综合提高了整个学习过程中的逼近能力和泛化能力。本文以Matlab7为平台设计了计算过程和具体的实现方式,还以分布于元宝露天矿区的6眼监测并为例,分别采用基于遗传算法的BP神经网络模型和BP神经网络模型对研究区的地下水位进行了短期预测,从对比分析的结果来看基于遗传算法的BP神经网络模型明显优于BP神经网络模型,是一种预测短期地下水位比较理想的预测模型。

Application of BP Neural Network Model to Groundwater Dynamic Predication Based on Genetic Algorithms

Aiming at the problems such that the prediction precision is low,the solutions are easy to converge to the local minimum points and the convergence speed is slow in the groundwater dynamic prediction with BP neural network,the groundwater predication model based on the GA-based BP neural network is presented.The calculation process and its realization ways are carded out based on the Matlab7.In the case study on 6 monitoring wells in Yuanbaoshan strip mining area,the GA-based BP neural network model and the BP neural network model are used separately to make a short-term predication for the groundwater level in the study area.According to the contrast results the GA-based model is obviously much better than another one,and it is really an optimal model for the short-term predication of groundwater level.