Groundwater quality evaluation using hybrid model of the multi-layer perceptron combined with neural-evolutionary regression techniques: case study of Shiraz plain

• Published in: Stochastic environmental research and risk assessment Vol. 37; no. 8; pp. 2961 - 2976
• Main Authors: Moayedi, Hossein, Salari, Marjan, Dehrashid, Atefeh Ahmadi, Le, Binh Nguyen
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2023; Springer Nature B.V
• Subjects: Accuracy; Aquatic Pollution; Artificial neural networks; Calcium ions; Chemistry and Earth Sciences; Computational Intelligence; Computer Science; Earth and Environmental Science; Earth Sciences; Environment; Groundwater; Groundwater data; Groundwater quality; Heuristic methods; Hydrologic data; Magnesium; Math. Appl. in Environmental Science; Multilayer perceptrons; Multilayers; Neural networks; Optimization; Original Paper; Physics; Prediction models; Probability Theory and Stochastic Processes; Quality assessment; Resource management; Statistics for Engineering; Swarm intelligence; Waste Water Technology; Water Management; Water Pollution Control; Water quality; Water resources management; Prediction performance; Artificial bee colony (ABC); Grey wolf optimizer (GWO); Groundwater quality; Artificial neural network (ANN); Harris hawks optimization (HHO); Shiraz plain
• ISSN: 1436-3240; 1436-3259
• DOI: 10.1007/s00477-023-02429-w

In recent decades, qualitative and quantitative assessments of groundwater sources reveal that efficient and accurate optimization approaches may assist in solving the multiple problems in evaluating groundwater quality. Hybrid models have been accepted and used in recent decades as a potentially useful approach for modeling water resource management processes in many different fields. Combined prediction models have more accurate outcomes than conventional methods. For this objective, three optimization meta-heuristic approaches, including Grey Wolf Optimizer (GWO), Harris Hawks Optimization (HHO), and Artificial Bee Colony (ABC), as well as intelligent models of Artificial Neural Networks (ANN), were employed to mimic groundwater quality. The input variables were Cl, SO 4 2− , HCO 3 − , Na + , Mg 2+ , Ca 2+ , Na percent, K + , pH, and total hardness (TH) as one of the water’s necessary quality factors for drinking/irrigation was output. To reach this purpose, the data on groundwater quality for the Shiraz plain were employed for a period of 16 years (2002–2018). As a result, for the training RMSE and R 2 databases, the estimated accuracy indices for the suggested hybrid HHO-ANN, GWO-ANN, and ABC-ANN models were (0.03907, 0.00427 , 0.1078) and (0.99258, 0.99991 , 0.94451), respectively, also for the testing RMSE and R 2 databases, these models were determined to be (0.03592, 0.00365, and 0.11944) and (0.99416, 0.99995, and 0.92628), respectively, for the testing datasets. Finally, the outcomes illustrated the high accuracy and capability of the GWO-ANN approach in simulating and appraising the quality of groundwater.