Scholarlink

<p>Accurate prediction of groundwater levels (GWL) is crucial for effective resource management and addressing challenges such as water scarcity and aquifer sustainability. This study aims to develop and evaluate a novel ensemble framework that integrates multiple machine learning (ML) models, including support vector regression (SVR), adaptive neuro-fuzzy inference systems (ANFIS), and long short-term memory (LSTM) networks, for robust GWL fluctuation prediction. The primary objectives are to (1) leverage the complementary strengths of these diverse ML techniques, (2) optimize their contributions through particle swarm optimization (PSO), and (3) validate the framework's performance across multiple regions in the Birjand watershed, Iran. The proposed method combines the time-series analysis capabilities of LSTM, the generalization power of SVR, and the non-linear mapping ability of ANFIS, achieving superior prediction accuracy. By employing PSO for dynamic weighting, the ensemble dynamically emphasizes the most effective models, resulting in statistically significant improvements over individual methods for metrics such as RMSE, MAE, and the inequality coefficient. Applied to four zones, the framework demonstrates consistent outperformance, with the best results achieved when considering GWL data from two prior months. The ensemble approach not only enhances prediction reliability but also addresses limitations in isolated modeling approaches, offering a scalable and adaptive solution for sustainable groundwater management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Intelligent Combination of Machine Learning Approaches for Groundwater Fluctuations Prediction

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Abbasali Rezapour,
  • Mostafa Sabzekar

摘要

Accurate prediction of groundwater levels (GWL) is crucial for effective resource management and addressing challenges such as water scarcity and aquifer sustainability. This study aims to develop and evaluate a novel ensemble framework that integrates multiple machine learning (ML) models, including support vector regression (SVR), adaptive neuro-fuzzy inference systems (ANFIS), and long short-term memory (LSTM) networks, for robust GWL fluctuation prediction. The primary objectives are to (1) leverage the complementary strengths of these diverse ML techniques, (2) optimize their contributions through particle swarm optimization (PSO), and (3) validate the framework's performance across multiple regions in the Birjand watershed, Iran. The proposed method combines the time-series analysis capabilities of LSTM, the generalization power of SVR, and the non-linear mapping ability of ANFIS, achieving superior prediction accuracy. By employing PSO for dynamic weighting, the ensemble dynamically emphasizes the most effective models, resulting in statistically significant improvements over individual methods for metrics such as RMSE, MAE, and the inequality coefficient. Applied to four zones, the framework demonstrates consistent outperformance, with the best results achieved when considering GWL data from two prior months. The ensemble approach not only enhances prediction reliability but also addresses limitations in isolated modeling approaches, offering a scalable and adaptive solution for sustainable groundwater management.