Loading...

Predicting Groundwater Level Changes in the Varamin Plain Using Machine Learning Models and Optimization Algorithms

Azizi, Hamed | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57656 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Khorashadizadeh, Farkhondeh
  7. Abstract:
  8. Predicting groundwater level fluctuations is essential for effective water resource management and environmental preservation. This study examines and compares two modeling techniques, namely Wavelet Neural Networks (WNN) and Artificial Neural Networks (ANN), for predicting fluctuations in the aquifer levels in the Varamin plain. Traditional hydrological models often struggle to detect nonlinear relationships present in hydrological data, which may lead to inconsistencies in predictions. The research focuses on groundwater level fluctuations in the Varamin plain over a 30-year period, from October 1989 to September 2018. Statistical methods such as mean, median, and variance calculations are used to analyze data dispersion and correlation. Additionally, outlier data is removed, and normalization is performed to ensure the validity of the input data for predictive models. The significance of these steps is particularly highlighted in the final accuracy of the predictive models and the identification of long-term trends in groundwater levels. In the second part, the use of machine learning models, particularly Artificial Neural Networks (ANN) and Wavelet Neural Networks (WNN), is explored as powerful tools for predicting groundwater levels. These models are utilized due to their ability to analyze complex and nonlinear relationships among variables, with the wavelet neural network yielding better results than the artificial neural network, achieving an MSE of 2.2429×10^(-4). Subsequently, optimization algorithms such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Firefly Algorithm (FA), and Imperialist Competitive Algorithm (ICA) are applied. Although all the optimization algorithms used in this research have managed to reduce the final error of the combined model, the analysis of the obtained data reveals that this reduction occurs only in the sixth decimal place. Furthermore, the results of all employed methods are completely identical up to five decimal places. This minor difference is insufficient to justify the use of optimization models, meaning these models will have no significant practical or applicative effect on the final predictions. Ultimately, this thesis provides practical solutions for the optimal management of groundwater resources in arid and semi-arid regions. The introduction of machine learning models and optimization algorithms represents a novel approach to water resource management, and the proposed recommendations can help address existing challenges regarding the sustainable exploitation of groundwater resources. The findings of this study could serve as a basis for effective policy-making in water resource management and improve the accuracy of hydrological predictions in future studies
  9. Keywords:
  10. Groundwater Level Prediction ; Machine Learning ; Neural Network ; Wavelet Analysis ; Optimization Algorithms ; Varamin Plain

 Digital Object List

 Bookmark

No TOC