Groundwater level forecasting using soft computing techniques

Abstract

Knowledge of groundwater level is very important in studies dealing with utilization and management of groundwater supply. Earlier studies have reported that ELM performs better than SVM for groundwater level prediction. This has been verified by comparing the prediction of groundwater levels at six locations in the district of Vizianagaram, Andhra Pradesh, using ANN, GP, SVM and ELM. Based on the comparison, it is observed that the performance of ELM is the best compared to other models. ELM is capable of predicting the nonlinear behavior of the groundwater levels. SVM performs better than GP and ANN. The performance of GP and ANN is analogous. Furthermore, an attempt has been made to enhance the performance of SVM by using SVM hybrid models such as SVM-QPSO and SVM-RBF, and the same has been compared with SVM and ELM. Results indicate that the performance of SVM-QPSO is far better compared to the performance of SVM and SVM-RBF. Moreover, performance of ELM is observed to be the best, but on some occasions, SVM-QPSO performs on par with ELM.