Predicting flocculant dosage in the drinking water treatment process using Elman neural network

Area of Science:

• Environmental Engineering
• Artificial Intelligence in Water Treatment
• Water Quality Management

Background:

• Accurate flocculant dosage is critical for public health in drinking water treatment.
• Water quality complexity and flocculation dynamics present significant prediction challenges.
• Existing models often struggle to capture the dynamic nature of water parameters.

Purpose of the Study:

• To develop and evaluate an artificial intelligence model for predicting drinking water flocculant dosage.
• To compare the performance of an Elman neural network (ENN) against other established models.
• To explore the practical application of advanced AI in waterworks operations.

Main Methods:

• Developed four models: Multiple Linear Regression (MLR), Radial Basis Function Neural Network (RBFNN), Least Squares Support Vector Machine (LSSVM), and Elman Neural Network (ENN).
• Incorporated a mixed data term (long-term and short-term) to capture water quality's periodic and time-varying characteristics.
• Utilized adaptive weight updates based on effluent turbidity and set values for model optimization.

Main Results:

• The Elman Neural Network (ENN) model demonstrated superior prediction performance over MLR, RBFNN, and LSSVM.
• ENN achieved significant improvements: 36.9% reduction in RMSE, 41.5% in MAPE, and 14.0% increase in R² compared to the best alternative (RBFNN).
• The ENN model resulted in more stable effluent turbidity in the sedimentation tank.

Conclusions:

• The Elman Neural Network (ENN) is a highly effective data intelligence tool for predicting drinking water flocculant dosage.
• The proposed mixed data term enhances the model's ability to handle complex water quality variations.
• ENN offers a reliable solution for optimizing water treatment processes and ensuring water safety.