Modeling of a Flotation Process in an Industrial WWTP Using Neural Networks

Objective: This study aimed to apply artificial neural networks (ANNs) to the flotation stage in the treatment of effluents from a beverage industry. Theoretical Framework: Dissolved air flotation (DAF) is widely used in the treatment of industrial effluents, particularly for the removal of contaminants such as phosphorus and organic matter. Due to the complexity and nonlinear behavior of the involved processes, artificial neural networks have emerged as effective tools for modeling and forecasting in multivariable and dynamic systems. Method: A feedforward ANN model was used, trained with the Levenberg-Marquardt algorithm and implemented in MATLAB. The network was configured with two hidden layers, 250 neurons, and a tansig activation function. Results and Discussion: The simulations showed that the predicted values of total phosphorus were consistent with those observed in industrial practice. The network training resulted in a mean squared error (MSE) of 1.87×10⁻² and a correlation coefficient (R) close to 1, indicating the model’s accuracy in relation to actual data. Research Implications: The application of ANNs proved effective for modeling and analyzing multivariable chemical processes under transient conditions, offering decision-making support for the control and optimization of the flotation process. Originality/Value: The main contribution of this study lies in demonstrating the ability of artificial neural networks (ANNs) to accurately model the behavior of a dissolved air flotation (DAF) system, even in the presence of significant dispersion in operational data. The robustness of the network in the face of variations in input variables confirms its potential to support process optimization, reduce contaminants in the final effluent, and mitigate eutrophication.