Water Quality Prediction Based on SSA-MIC-SMBO-ESN

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Water pollution poses significant risks to human activities and ecosystems.
• Proactive water quality prediction is crucial for effective water resource management.
• Existing prediction methods may be limited by data noise and inter-variable correlations.

Purpose of the Study:

• To develop accurate predictive models for key water quality indicators: dissolved oxygen (DO), permanganate index (CODMn), and total phosphorus (TP).
• To leverage the temporality of water quality data using Echo State Networks (ESN).
• To enhance prediction accuracy by incorporating data denoising and feature correlation analysis.

Main Methods:

• Data preprocessing included imputation of missing values and outlier correction using Z-score and linear trend methods.
• Singular Spectrum Analysis (SSA) was employed for denoising time-series water quality data.
• Maximum Information Coefficient (MIC) was used to identify strong correlations between water quality indices for feature selection.
• Multi-feature water quality prediction models were built using offline and online learning algorithms of ESN.
• Hyperparameter optimization was performed using Sequential Model-Based Optimization (SMBO).

Main Results:

• The developed SSA-MIC-SMBO-Offline ESN and SSA-MIC-SMBO-Online ESN models demonstrated high accuracy in predicting DO, CODMn, and TP.
• Data denoising via SSA effectively improved model performance by mitigating noise interference.
• Feature selection based on MIC identified relevant indices, enhancing the predictive power of the ESN models.

Conclusions:

• The proposed ESN-based models offer a robust and accurate approach for water quality prediction.
• These models provide valuable tools for water management authorities to anticipate and respond to pollution events.
• The integration of SSA, MIC, and SMBO with ESN represents an effective strategy for time-series water quality forecasting.