Enhancing sustainability in a sequential batch reactor for wastewater treatment: soft sensor-based ammonia prediction with PSO-LSTM and random forest model

Abstract

Water quality monitoring in wastewater treatment is crucial due to its significant impact on environmental sustainability and public health. Small wastewater treatment facilities often encounter challenges such as limited resources, lack of advanced monitoring technologies, and difficulties in maintaining effluent quality standards. This study aims to address these challenges by proposing an integrated methodology that employs a particle swarm optimization-based long short-term memory (PSO-LSTM) model in conjunction with a random forest regression (RFR) model for the prediction of ammonia nitrogen (NH₃-N) concentrations. The methodology integrates the strengths of both models to enhance predictive accuracy and stability. The PSO algorithm optimizes hyperparameters of the LSTM, effectively searching the solution space to balance exploration and exploitation. This optimization facilitates the handling of complex non-linear data patterns while improving computational efficiency. The integrated PSO-LSTM-RFR approach demonstrated superior performance on the test set, achieving a coefficient of determination (R²) of 0.9105, a root mean square error (RMSE) of 2.9706, and a mean absolute error (MAE) of 2.5430. The results confirm that the proposed methodology significantly enhances predictive performance due to the complementary advantages of the models. The synergy between RFR and LSTM allows for effective handling of non-linear relationships and sequential data trends, leading to more accurate predictions. The findings underscore the potential of the integrated PSO-LSTM-RFR model to enhance predictive efficacy in wastewater treatment processes, ultimately contributing to more effective water quality management strategies.