NEURAL NETWORK-BASED PREDICTION OF DISSOLVED SALT CONCENTRATION IN REVERSE OSMOSIS PERMIAT

Traditional methods for monitoring the quality of purified water are often based on periodic measurements and simplified models, which do not allow for timely prediction of potential deterioration in permeate quality.Under these conditions, the application of artificial intelligence methods, particularly neural networks, becomes relevant, as they are capable of analyzing large datasets and modeling complex nonlinear relationships between technological parameters.The aim of this work was to apply artificial intelligence methods, specifically neural networks, to improve drinking and process water purification systems, particularly membrane systems, which can model complex dependencies between numerous input parameters and the output result.In this study, a theoretical analysis was conducted, and three models were developed based on different neural network architectures-a multilayer perceptron, a long short-term memory recurrent network, and a convolutional neural network.These models enable the prediction of dissolved salt concentrations in permeate based on five technological input parameters: applied pressure, feed solute concentration, flow rate, temperature, and membrane permeability index.For each model, the training dynamics, prediction accuracy, and correspondence to actual values were analyzed.A comprehensive evaluation of the effectiveness of the three implemented models-the multilayer perceptron, the long short-term memory recurrent network, and the convolutional neural network-was performed.After training each model, the mean absolute error between the predicted value and the true (reference) value was calculated.For visual error analysis, epochby-epoch plots were constructed for each model.The analysis of the obtained results showed that all three models are capable of generating approximate predictions; however, their accuracy and stability vary.The convolutional neural network proved to be the most suitable for predicting salt concentration in permeate.Its integration into an automated water purification control system may significantly enhance monitoring efficiency and real-time decision-making.