Using deep learning models to predict the electrical conductivity of the influent in a wastewater treatment plant

Abstract

Nowadays, human population face increasing water pollution problems, so treating and managing this resource is crucial. Wastewater Treatment Plants (WWTPs) provide essential services for human life since they treat wastewater and monitor its parameters to preserve water quality standards. One of these parameters is electrical conductivity, essential in quantifying water salinity levels. Therefore, this paper aims to forecast the influent conductivity in a WWTP for the next two timesteps. Hence, several experiments were conducted, considering the use of Transformers and Long Short-Term Memory (LSTMs) candidate models that were developed, tuned, and evaluated, utilising a recursive multi-step forecasting approach. The best candidate model was based on a Transformer architecture with encoding and obtained a RMSE of 155.2 μ S/cm.