|
Determining the best observation wells to predict groundwater depth using ANFIS based on different training algorithms
|
Abbas Sedghamiz *    |
| Assistant Professor, Department of Irrigation Technology, Collage of Agriculture and Natural Resources of Darab, Shiraz University, Iran, Email: sedghamiz@shirazu.ac.ir |
|
|
Abstract: (396 Views) |
| For the effective and optimal management of groundwater resources, accurate predictions considering all prevailing conditions in aquifers, particularly fluctuations in groundwater level and depth, are essential. The objective of this study, conducted in the Qotbabad region of Jahrom County, Fars Province, is to identify observation wells that provide the most reliable predictions of groundwater depth in other wells. To achieve this, the Adaptive Neuro-Fuzzy Inference System (ANFIS) was employed, in combination with various training algorithms including Hybrid, Genetic Algorithm (GA), and Particle Swarm Optimization (PSO). Groundwater depth data from seven observation wells across the plain were used, covering the period from October 2008 to September 2024. To evaluate model accuracy, statistical indices such as Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE) were utilized. Based on the results, observation well No. 2 was identified as the most accurate predictor for wells No. 1 and 4, while well No. 5 was identified as the least accurate predictor for wells No. 3 and 6. Additionally, observation well No. 4, classified as a moderately accurate predictor, demonstrated the best predictive performance for wells No. 2, 5, and 7, and ranked second-best for wells No. 1, 3, and 6. This consistent ranking as either the top or second-best predictor sets well No. 4 apart from the others. Among all wells, the strongest linear relationship between observed and predicted groundwater depths was obtained for well No. 4, with an average coefficient of determination (R²) of 0.9945 across the three training algorithms. Conversely, the weakest relationship was found for well No. 3, with an average R² of 0.7435. Overall, the Hybrid method proved to be the most accurate and the fastest to execute, whereas the Genetic Algorithm method, having the most execution time, exhibited the lowest predictive accuracy. |
Article number: 7 |
| Keywords: Groundwater, Adaptive Neouro- Fuzzy Inference System, Training Algorithme, Error Index |
|
|
Full-Text [PDF 2038 kb]
(124 Downloads)
|
|
Type of Study: Research |
Subject:
Special
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
