Application of LM and GDX algorithms in ANN modeling to Estimate the Groundwater Level Fluctuations

Abstract

In the present investigation prediction of groundwater (GW) level near Varanasi is done by Artificial Neural Network (ANN) model. ANN is one of the methods used for development of the prediction model based upon the working of the human brain. This study gives a perfect prediction trained with two training algorithms LM (Levenberg-Marquardt) and GDX (Adaptive Learning rate with back Propagation). Data of eight wells, Annual precipitation, maximum and minimum Temperature, Relative humidity are the parameters taken as input, while future groundwater levels as output. The model proficiency and precision were estimated dependent on the R (regression coefficient) and RMSE (root mean square error) values approach. Observed R and RMSE values for most of the wells were tending towards unity in the LM approach. This LM approach is useful when we are short in terms of data, and it is expected that this strategy will give a genuinely exact outcome for a long span of data under examination. The LM method is discovered to be appropriate for determining any forecast of water fluctuations when there is a limitation in data constraint. This technique also gives an accurate result when the river is taken as an input in the artificial neural network (ANN) and the other input parameters.