Tracking Health Risk-Based Resistivity in Treated Wastewater-Based on Pilot-System using Sensors, IoT, and 2nd-order Ensemble Machine Learning Algorithms

This study establishes the real-time monitoring and prediction of treated wastewater quality in Al-Hassa, Saudi Arabia, utilizing an integrated system of sensors, the Internet of Things (IoT), and machine learning. Targeting resistivity as a key indicator of salinity and purity, the research aims to advance water resource management through technological innovation. The methodology employs IoT-enabled sensors to gather data on temperature, Oxidation Reduction Potential (ORP), Electrical Conductivity (EC), and Total Dissolved Solids (TDS). Dependency analysis identifies crucial relationships between these variables. Subsequent model validation compares various neural network architectures Narrow Neural Networks (NNN), Wide Neural Networks (WNN), Bilateral Neural Networks (BNN), and Simple Average Ensemble (SAE) based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and prediction speed. BNN outstrips other models, with an
$\text{RMSE}=0.0008$
and an
$\text{MAE}=0.0003$
, suggesting superior accuracy, and a prediction speed of 33,000 observations per second, showcasing its computational efficacy. In contrast, NNN and WNN models demonstrate
$\text{RMSE }=0.0009$
and 0.0012 respectively, while SAE presents an intermediate
$\text{RMSE }=0.000967$
. These results underscore the effectiveness of BNN in the accurate and efficient prediction of resistivity, providing a robust solution for automated water quality monitoring. The study's findings hold significant promise for enhancing environmental monitoring and offer a scalable blueprint for water-scarce regions on a global scale.