A water quality prediction model based on signal decomposition and ensemble deep learning techniques.

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Accurate water quality prediction is essential for effective water resource protection.
• Dissolved oxygen (DO) is a key indicator of river water quality and ecosystem health.
• Existing prediction models often struggle with the complexity of river water systems.

Purpose of the Study:

• To develop and validate a hybrid model for predicting river water quality, specifically dissolved oxygen (DO).
• To enhance the accuracy and generalizability of water quality forecasting in complex river environments.
• To integrate advanced signal decomposition and deep learning techniques for improved prediction.

Main Methods:

• Utilized Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to decompose DO time series into Internal Mode Functions (IMFs).
• Applied Multi-Scale Fuzzy Entropy (MFE) to analyze the entropy of each IMF component.
• Employed Time-Varying Filtered Empirical Mode Decomposition (TVFEMD) for feature extraction in high-frequency subsequences.
• Integrated Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) neural networks for predicting low- and high-frequency components.

Main Results:

• The proposed hybrid model demonstrated superior prediction accuracy and generalizability compared to single models and other ensemble approaches.
• Validation on Xinlian section of Fuhe River and Chucha section of Ganjiang River confirmed the model's effectiveness.
• The fusion of CEEMDAN, MFE, TVFEMD, SVM, and LSTM significantly improved dissolved oxygen prediction.

Conclusions:

• The developed hybrid model offers a robust and accurate method for forecasting river water quality.
• This approach is particularly effective for predicting water quality in complex river systems.
• The findings provide a valuable tool for water resource management and ecological health monitoring.