LSTM-based Modelling for Coagulant Dosage Prediction in Wastewater Treatment Plant

Abstract

The dosage of coagulant plays a critical role in ensuring effluent quality, however the complexity of the coagulant chemical theory and affected by many factors (turbidity, pH, conductivity, flow rate, etc.) that it is difficult to determine the optimal dosage effectively. Optimization of coagulant dosage in wastewater treatment is becoming more critical as water quality standards become increasingly stringent. In previous studies, usually build a prediction model only use current water quality parameters that the water quality parameters of previous time sequence were ignored, result in the prediction accuracy is not satisfactory. In this paper, not only current water quality parameters have been taken into account during the modeling, but also historical time-series water quality feature data also considered. For this purpose, a long short term memory (LSTM) model is applied, that is effectively solved the problem of long-term dependencies of recurrent neural network. We collected real sewage treatment plant data for experiments, thorough empirical studies based upon the dataset, and we use R2, RMSE and MAPE as evaluation metrics, experimental result demonstrate that based on LSTM algorithm model can outperform state-of-the-art prediction accuracy compared other algorithm model.