Water quality forecasting based on data decomposition, fuzzy clustering and deep learning neural network

Area of Science:

• Environmental Science
• Data Science
• Hydrology

Background:

• Accurate water quality forecasting is crucial for public health and water resource management.
• Existing models often face challenges in capturing complex temporal dynamics in water quality data.

Purpose of the Study:

• To develop a novel hybrid model for enhanced water quality forecasting accuracy.
• To improve upon existing forecasting methods by integrating data decomposition and advanced machine learning techniques.

Main Methods:

• The proposed model utilizes Empirical Wavelet Transform for data decomposition.
• Fuzzy C-means clustering is employed to recombine decomposed subseries.
• Bidirectional Gated Recurrent Unit (BiGRU) networks are applied for prediction on clustered series.

Main Results:

• The hybrid model achieved a high forecast accuracy, with an average Mean Absolute Percentage Error (MAPE) of 4.59% for 7-day ahead predictions.
• The model demonstrated superior performance compared to other benchmark models, including a 32.86% average reduction in MAPE compared to a single BiGRU model.
• The model effectively forecasted six different water quality parameters for Poyang Lake.

Conclusions:

• The novel hybrid model offers a significant advancement in water quality forecasting.
• The integration of data decomposition, clustering, and BiGRU provides a robust framework for accurate environmental prediction.
• The proposed method is effective for real-world water quality forecasting applications.