Employing artificial neural network for accurate modeling, simulation and performance analysis of an RO-based desalination process

In this paper, we implement sustainable soft computing via application of Artificial Neural Networks (ANN) to predict the desalination plant output more accurately. The proposed ANN models have been trained with a backpropagation (BP) algorithm. Motivated by the literature, an exhaustive modeling has been performed using an extended list of six ANN modeling parameters that include number of hidden layers and their nodes, activation and training functions, dataset fractional allocation, and error analysis functions. For this, 4 suitable experimental inputs and permeate flux as output have been considered for the modeling investigation. The proposed models have been simulated in the MATLAB/Simulink platform and compared with literature models, based on the regression coefficient (R²), errors, and iterations. An investigation of 127 trials leads to the optimal selection of the parameters that achieve higher optimal model performance (R² = 99.4%, Error = 0.003) than the existing models in the literature. We have observed that choice of softmax-purelin function (activation function for hidden layer-output layer), two hidden layers with 20 nodes each, Levenberg–Marquardt training function, dataset divisions (80% training: 10% validation: 10% testing) and the mean square error (MSE) demonstrate best model performance. Such an efficient and sustainable soft computing system modeling and simulation exemplifies an indispensable method for supporting the desalination design and engineering and assists in an efficient and robust control of the desalination plants.