Predictive modelling and optimization of an airlift bioreactor for selenite removal from wastewater using artificial neural networks and particle swarm optimization

Selenite (Se4+) is the most toxic of all the oxyanion forms of selenium. In this study, a feed forward back propagation (BP) based artificial neural network (ANN) model was developed for a fungal pelleted airlift bioreactor (ALR) system treating selenite-laden wastewater. The performance of the bioreactor, i.e., selenite removal efficiency (REselenite) (%) was predicted through two input parameters, namely, the influent selenite concentration (ICselenite) (10 mg/L - 60 mg/L) and hydraulic retention time (HRT) (24 h - 72 h). After training and testing with 96 sets of data points using the Levenberg-Marquardt algorithm, a multi-layer perceptron model (2-10-1) was established. High values of the correlation coefficient (0.96 <= R <= 0.98), along with low root mean square error (1.72 <= RMSE <= 2.81) and mean absolute percentage error (1.67 <= MAPE <= 2.67), clearly demonstrate the accuracy of the ANN model (> 96%) when compared to the experimental data. To ensure an efficient and economically feasible operation of the ALR, the process parameters were optimized using the particle swarm optimization (PSO) algorithm coupled with the neural model. The REselenite was maximized while minimizing the HRT for a preferably higher range of ICselenite. Thus, the most favourable optimum conditions were suggested as: ICselenite - 50.45 mg/L and HRT - 24 h, resulting in REselenite of 69.4%. Overall, it can be inferred that ANN models can successfully substitute knowledge-based models to predict the REselenite in an ALR, and the process parameters can be effectively optimized using PSO.