Development of virtual sensors using multi-plant data for wastewater treatment applications: a case study on nitrogen prediction in effluents

Abstract This paper presents the implementation of a virtual sensor for real-time prediction of total nitrogen in the effluent of two wastewater treatment plants. The virtual sensor is designed to complement or temporarily replace traditional sensors, which are prone to failure and require regular maintenance. By applying several regression techniques, including recursive least squares, decision trees, and machine learning algorithms, the study identifies the most accurate model for predicting total nitrogen at the outlet. The results demonstrate that the virtual sensor can reliably estimate the target variable using other measurements at the plant, providing a cost-effective and robust solution to improve operations. The study highlights the potential of regression-based virtual sensors to optimize plant management without the need for significant investment in additional infrastructure, thus creating a viable methodology for use as well as a tool capable of detecting anomalies present in real measurements.