A novel hybrid model based on two-stage data processing and machine learning for forecasting chlorophyll-a...

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Accurate chlorophyll-a (Chl-a) prediction is vital for early algal bloom detection in reservoirs.
• Multivariate time series prediction of Chl-a is challenging due to complex aquatic environment interactions and non-stationary water quality data.

Purpose of the Study:

• To propose a novel prediction model, SGMD-KPCA-BiLSTM (SKB), for enhanced Chl-a concentration prediction.
• To address the challenges of nonlinear relationships and non-stationary data in water quality monitoring.
• To improve the accuracy and efficiency of algal bloom early warning systems.

Main Methods:

• Combined Symplectic Geometry Mode Decomposition (SGMD) and Kernel Principal Component Analysis (KPCA) for optimal nonlinear data subset selection.
• Utilized a Bidirectional Long Short-Term Memory (BiLSTM) model for time series prediction.
• Optimized model hyperparameters using the Sparrow Search Algorithm (SSA).

Main Results:

• The SKB model demonstrated superior performance in Chl-a prediction, achieving R² = 96.19%, RMSE = 1.05, MAE = 0.65, and MAPE = 0.08.
• Significantly reduced prediction errors compared to single and hybrid models like BP, SVR, LSTM, CNN-LSTM, BiLSTM, SGMD-LSTM, SGMD-KPCA-LSTM, and SGMD-BiLSTM.
• Showcased reliable multi-step predictive capabilities, with performance declining gradually with increased prediction time steps.

Conclusions:

• The SKB model offers a robust solution for predicting non-smooth and nonlinear Chl-a sequences from online monitoring systems.
• This approach provides a potential strategy for reservoir eutrophication control and prevention.
• Presents an innovative method for advancing water quality prediction and management.