Development of machine learning model for the sustainable phycoremediation of municipal wastewater

This study explored the application of artificial intelligence/machine learning (AI / ML) techniques to model the interaction of 19 water quality parameters with respect to the concentrations of biogenic CO2 emission and Chemical Oxygen Demands (COD) as the target outputs, under a bioreactor setup with algae-activated sludge ratio of 1:5. The results showed that the ideal AI / ML model had a typology of-19-20-2 [multi-input multi output (MIMO) feed-forwards neural network], learned under Levenberg-Marquart backpropagation algorithm and with transfer function log-sigmoid. During the training stage, the overall coefficient of determination (R2) and mean squared error (MSE) were 0.997 and 94.46 mg L−1, respectively. After the validation, the obtained values for CO2 emission prediction were R2 = 0.992 and root mean square error (RMSE) = 10.72 mg L−1, while for COD prediction, the values were R2 = 0.991 and root mean square error (RMSE) = 1.125 mg L−1. These results demonstrated that the prediction values fit the actual values well with only small errors and high accuracy. Moreover, Principal Component Analysis found that Total Suspended Solids and Mixed Liquor Volatile Suspended Solids were highly correlated with COD, while pH, Total Phosphorus and NH4+-N were somehow close to CO2 emissions. This is the first known application of AI/ML in a phycoremediation system that addresses greenhouse gas (GHG) and water quality issues simultaneously. This pioneering work bridges the gap between wastewater quality and GHG emissions, significantly advancing the field of municipal wastewater treatment and contributing to sustainable development.