Scholarlink

<p>Forecasting water quality indicators (WQIs) is of high importance for pollution management and sustainability efficiently. Long-term forecasting often faces difficulties due to complex data trends and insufficient feature extraction, leading to considerable environmental impacts. This study introduces a sophisticated hybrid deep learning framework, named VMD-CNN-GRU, which integrates Variational Model Decomposition (VMD), Convolutional Neural Networks (CNN), and Gated Recurrent Units (GRUs) to address current challenges. Such an integrated farmwork boosts the accuracy of water quality forecasts for short, mid, and long-term periods. In addition to the hybrid model, two separate deep learning models (CNN and GRU) and one machine learning model (Support Vector Regression) were developed to predict two key water quality indicators: dissolved oxygen (DO) and chlorophyll-a (Chl-a). A total of 48 models with diverse structures were evaluated, including 24 models for Chl-a forecasting and 24 models for DO forecasting within the same area. Although data-driven models are commonly employed for short- and mid-term predictions, there remains a notable gap in the area of long-term hybrid modeling. The VMD-CNN-GRU framework successfully tackles this issue by combining decomposition with deep learning to capture nonlinear and nonstationary behaviors. The model exhibited exceptional performance relative to other machine learning and deep learning techniques, achieving margins of 34% over SVR, 31% over CNN, and 30% over GRU, thereby highlighting the effectiveness of VMD in long-term water quality forecasting.</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 Integrated VMD-CNN-GRU Framework for Multi-Horizon Forecasting of Water Quality Dynamics

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • Mojtaba Sabagh Torkan,
  • Abbas Hekmatiyan,
  • Mohammad Hashemzadeh,
  • Amin Karbasi Ahvazi,
  • Farshad Hesamfar,
  • Mohammad G. Zamani

摘要

Forecasting water quality indicators (WQIs) is of high importance for pollution management and sustainability efficiently. Long-term forecasting often faces difficulties due to complex data trends and insufficient feature extraction, leading to considerable environmental impacts. This study introduces a sophisticated hybrid deep learning framework, named VMD-CNN-GRU, which integrates Variational Model Decomposition (VMD), Convolutional Neural Networks (CNN), and Gated Recurrent Units (GRUs) to address current challenges. Such an integrated farmwork boosts the accuracy of water quality forecasts for short, mid, and long-term periods. In addition to the hybrid model, two separate deep learning models (CNN and GRU) and one machine learning model (Support Vector Regression) were developed to predict two key water quality indicators: dissolved oxygen (DO) and chlorophyll-a (Chl-a). A total of 48 models with diverse structures were evaluated, including 24 models for Chl-a forecasting and 24 models for DO forecasting within the same area. Although data-driven models are commonly employed for short- and mid-term predictions, there remains a notable gap in the area of long-term hybrid modeling. The VMD-CNN-GRU framework successfully tackles this issue by combining decomposition with deep learning to capture nonlinear and nonstationary behaviors. The model exhibited exceptional performance relative to other machine learning and deep learning techniques, achieving margins of 34% over SVR, 31% over CNN, and 30% over GRU, thereby highlighting the effectiveness of VMD in long-term water quality forecasting.