Application of RBF Neural Network for Groundwater Dynamic Prediction

曹文洁,胡冰,韩良跃,梁秀娟,肖长来

DOI: https://doi.org/10.13928/j.cnki.wrahe.2018.02.007

2018-01-01

Abstract:In order to predict dynamically groundwater level,the neural network model is applied to the construction of groundwater depth prediction model.Taking Changchun city as an example,the dynamic process of buried depth is simulated by radial basis function(RBF) and reverse propagation(BP) algorithm model optimization parameters within 84 sets of data from 2006 to 2012 as training sample and 36 sets of data from 2013 to 2015 as test sample,in full of approximation convergence cappcity of RBF nerve network.The two models simulate the measured buried depth dynamic process,and compare the performance.The results indicate that following parameters of the RBF neural network model and BP neural network model:the root mean square errors are 0.10 and 0.43,respectively.The maximum absolute errors are 0.44 m and 0.61 m,respectively,and the maximum relative errors are 14.60％ and 27.17％ respectively;groundwater level of Changchun city has obvious cyclic and large variation after 2015 with obvious seasonal characteristics,and the maximum depth is 5.1 m in rainless preriod but the minimum depth is 1.62 rn in wet season,obviously,the RBF model has better nonlinear mapping ability and prediction accuracy.The model can be used to predict the dynamic data of similar.