Comparative efficiency of different artificial intelligence based models for predicting density dependent saltwater intrusion processes in coastal aquifers and saltwater intrusion management utilizing the best performing model

Roy, Dilip Kumar, and Datta, Bithin (2018) Comparative efficiency of different artificial intelligence based models for predicting density dependent saltwater intrusion processes in coastal aquifers and saltwater intrusion management utilizing the best performing model. Desalination and Water Treatment, 105. pp. 160-180.

[img] PDF (Published Version) - Published Version
Restricted to Repository staff only

View at Publisher Website: https://doi.org/10.5004/dwt.2018.21982
 
1
2


Abstract

Artificial intelligence based data driven models are useful tools in approximating density dependent coupled flow and salt transport processes in coastal aquifer systems. These emulators often serve as computationally efficient substitutes of rigorous numerical simulation models within a linked simulation-optimization (S/O) methodology. In this study, fuzzy c-means clustering based fuzzy inference systems (FIS) and adaptive network based fuzzy inference systems (ANFIS) are proposed to approximate the physical processes of an illustrative coastal aquifer system. FIS and ANFIS based models are also utilized to identify the most influential input variables in predicting salinity concentrations at three monitoring locations (C1, C2, and C3). Solution results obtained are compared with those obtained using a genetic programming (GP) based modelling approach. Performance evaluation results show that the developed FIS and ANFIS models perform equally well on training and testing datasets. It is also demonstrated that performances of both the FIS and ANFIS models are better than that of GP based models. Root mean square error (RMSE) and mean absolute percentage relative error (MAPRE) values obtained by using GP are larger than those obtained by both ANFIS and FIS models. The maximum prediction errors of GP represented by RMSE and MAPRE at the three monitoring locations are 22.43 mg/l and 2.84% respectively. On the other hand, GP results in lowest values of correlation coefficient (0.96) and Nash-Sutcliffe efficiency coefficient (0.91). FIS model outperforms both ANFIS and GP models in terms of a more detailed comparison criteria, including computation time and model complexity. FIS requires only 0.65 min for training in order to predict salinity concentrations at all three locations C1, C2, and C3. For the similar purpose, ANFIS and GP require 17.35 min and 276.5 min, respectively. Therefore, FIS model can be successfully applied as a computationally efficient substitute of complex numerical simulation models for predicting coupled flow and salt transport processes. Such applications can be very useful in developing a computationally feasible linked S/O methodology for regional scale management of saltwater intrusion in coastal aquifers. Results of the management model using the best performing global FIS model indicates that FIS model provides acceptable, accurate, and reliable groundwater extraction patterns to limit the saltwater concentrations within the pre-specified maximum allowable limits.

Item ID: 55756
Item Type: Article (Research - C1)
ISSN: 1944-3986
Keywords: Coastal aquifer, Saltwater intrusion, C-means clustering, Genetic programming, Fuzzy inference system, Adaptive neuro-fuzzy inference system, Management model
Date Deposited: 03 Oct 2018 09:17
FoR Codes: 09 ENGINEERING > 0905 Civil Engineering > 090509 Water Resources Engineering @ 100%
SEO Codes: 96 ENVIRONMENT > 9609 Land and Water Management > 960999 Land and Water Management of Environments not elsewhere classified @ 100%
96 ENVIRONMENT > 9609 Land and Water Management > 960999 Land and Water Management of Environments not elsewhere classified @ 100%
Downloads: Total: 2
More Statistics

Actions (Repository Staff Only)

Item Control Page Item Control Page