Modelling and optimisation of activated sludge process using artificial neural networks and genetic algorithms

Mathematical modelling of the activated sludge process (ASP) was performed using multi-layer perceptron neural networks (MLP-ANN) to predict effluent water quality parameters and multi-objective genetic algorithm (MOGA) was employed to optimise influent water quality parameters so that the concentration of contaminants in the effluent stream is minimised. The study area selected was located in a central district of a southern state of India. The effluent parameters to be investigated and optimised are pH, suspended solids (SS) and biochemical oxygen demand (BOD) and oil and grease (O&G). The model was evaluated based on the statistical parameters of the correlation coefficient R and the mean square error (MSE). MATLAB R2019a was used for the modelling and optimisation study. It has been found that effluent pH, SS and BOD were predicted with an overall R of 0.9207 and an MSE of 0.0091. During optimisation of influent parameters, it was found that optimum values of the decision variables \mathrm{pH_{Inf}} lie between 6–8, optimum values of \mathrm{SS_{Inf}} lie between 68–380 mg L^{-1} , optimum values of \mathrm{BOD_{Inf}} lie between 155–692 mg L^{-1} and optimum values of O& \mathrm{G_{Inf}} lie between 8–45 mg L^{-1} when the objective functions were minimised simultaneously.