Estimating the incubated river water quality indicator based on machine learning and deep learning paradigms: BOD5...

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Machine Learning Applications

Background:

• Biochemical oxygen demand (BOD5) is a key indicator of river water quality and a significant pollutant impacting aquatic ecosystems.
• Accurate and cost-effective prediction of BOD5 is essential for effective water resource management.
• Machine learning and deep learning approaches are increasingly employed for water quality prediction.

Purpose of the Study:

• To evaluate the predictive performance of standalone machine learning (Extreme Learning Machine, Support Vector Regression) and deep learning (Deep Echo State Network) models for BOD5 concentration.
• To investigate the efficacy of novel double-stage synthesis models integrating Wavelet Transformation (WT) with these standalone models.
• To compare the accuracy of these models using various input combinations and performance metrics.

Main Methods:

• Developed and assessed standalone models: Extreme Learning Machine (ELM), Support Vector Regression (SVR), and Deep Echo State Network (Deep ESN).
• Integrated Wavelet Transformation (WT) to create novel double-stage synthesis models: Wavelet-ELM, Wavelet-SVR, and Wavelet-Deep ESN.
• Evaluated models using five input associations comprising diverse water quantity and quality parameters.
• Assessed model performance with Coefficient of Determination (R²), Nash-Sutcliffe (NS) efficiency, and Root Mean Square Error (RMSE).

Main Results:

• Standalone and double-stage synthesis models showed varying degrees of predictive accuracy; double-stage models were not consistently superior.
• At Hwangji station, Support Vector Regression (SVR) with the 3rd distribution (NS = 0.915) and Wavelet-SVR with the 4th distribution (NS = 0.915) provided highly accurate BOD5 predictions.
• At Toilchun station, Wavelet-SVR with the 4th distribution (NS = 0.917) emerged as the most effective model for predicting BOD5 concentration.
• Both standalone and novel synthesis models demonstrated utility in managing water pollutants.

Conclusions:

• The study highlights the effectiveness of both standalone and wavelet-enhanced models in predicting BOD5 concentration, crucial for water quality assessment.
• While novel double-stage models offer potential, standalone models like SVR can also achieve high accuracy.
• The findings support the application of these advanced modeling techniques for efficient data administration and regulation of water pollutants in riverine ecosystems.
• Wavelet-SVR demonstrated superior performance in specific instances, indicating its value for water quality monitoring.