Open AccessPredicting Hydraulic Conductivity (k) of Tropical Soils by using Artificial Neural Network (ANN)
Author(s) -
Dae-Sung Lim,
Prabir K. Kolay
Publication year - 2009
Publication title -
journal of civil engineering, science and technology
Language(s) - English
Resource type - Journals
ISSN - 2462-1382
DOI - 10.33736/jcest.63.2009
Subject(s) - conjugate gradient method , hydraulic conductivity , backpropagation , artificial neural network , rprop , richards equation , broyden–fletcher–goldfarb–shanno algorithm , soil water , infiltrometer , mathematics , soil science , algorithm , hydraulic head , computer science , artificial intelligence , geotechnical engineering , environmental science , engineering , types of artificial neural networks , computer network , asynchronous communication , time delay neural network
Hydraulic conductivity of tropical soils is very complex. Several hydraulic conductivity prediction methods have focused on laboratory and field tests, such as the Constant Head Test, Falling Head Test, Ring Infiltrometer, Instantaneous profile method and Test Basins. In the present study, Artificial Neural Network (ANN) has been used as a tool for predicting the hydraulic conductivity (k) of some tropical soils. ANN is potentially useful in situations where the underlying physical process relationships are not fully understood and well-suited in modeling dynamic systems on a real-time basis. The hydraulic conductivity of tropical soil can be predicted by using ANN, if the physical properties of the soil e.g., moisture content, specific gravity, void ratio etc. are known. This study demonstrates the comparison between the conventional estimation of k by using Shepard's equation for approximating k and the predicted k from ANN. A programme was written by using MATLAB 6.5.1 and eight different training algorithms, namely Resilient Backpropagation (rp), Levenberg-Marquardt algorithm (lm), Conjugate Gradient Polak-Ribiere algorithm (cgp), Scale Conjugate Gradient (scg), BFGS Quasi-Newton (bfg), Conjugate Gradient with Powell/Beale Restarts (cgb), Fletcher-Powell Conjugate Gradient (cgf), and One-step Secant (oss) have been compared to produce the best prediction of k. The result shows that the network trained with Resilient Backpropagation (rp) consistently produces the most accurate results with a value of R = 0.8493 and E2 = 0.7209.