AI-Based Water Quality Prediction for Full Automation of Water Treatment Plant Operations

Abstract Submission: The increasing significance of water supply, driven by global climate changes, population growth, and infrastructure development, has led to a heightened focus on optimizing water treatment processes. Recent advancements have seen the application of AI technologies in predicting water quality, particularly for controlling chlorine levels in treatment plants that utilize pre- or intermediate chlorination. Despite this progress, the integration of AI in water treatment facilities remains challenging due to technical and mechanical limitations. This study presents the development of an AI-based model designed to predict residual chlorine concentration at the sedimentation basin's outlet, with the goal of automating chlorine control processes. The model leverages real-time data on water quantity and quality, using variables such as the residual chlorine concentration at the basin's inlet, turbidity, pH, water temperature, electrical conductivity, raw water inflow, alkalinity, and ammonia (NH3) as inputs, while predicting the residual chlorine concentration at the outlet. Given the frequent occurrence of outliers and missing data due to mechanical issues in treatment plants, the study includes a comprehensive preprocessing method. This method automates the detection and removal of outliers using Local Outlier Factor (LOF) and imputes missing values through a combination of the Mann-Kendall Test and ARIMA model. The model development employed multiple regression, random forest, and Long Short-Term Memory (LSTM) networks, with hyper-parameter optimization for the latter two using genetic algorithms to ensure peak predictive accuracy. The final model was selected based on performance evaluation. The fully algorithmic approach presented here is anticipated to significantly enhance the operational efficiency of water treatment plants when implemented in real-world conditions.

Learning Objectives/Expected Outcome (Optional) :