Deep Learning for Ultrafiltration Membrane Prediction in Drinking Water Treatment

Ultrafiltration membrane (UFM) technology offers an efficient filtration solution for purifying underground water sources into potable drinking water. However, membrane fouling is one of main problems in this technology. Developing membrane filtration models is imperative for predicting and managing fouling occurrences during the filtration process. In this study, two deep learning models, namely long-short term memory (LSTM) and a hybrid GRU-LSTM, were employed to forecast transmembrane pressure (TMP) in UFM systems. Leveraging the capacity of deep learning LSTM to manage extensive dependencies inherent in long-range data, a dataset of 6686 observations was utilized. The results revealed that the hybrid gated recurrent unit long-short term memory (GRU-LSTM) model outperformed the LSTM model, achieving an R2 value of 97% compared to LSTM’s 92.5%. This underscores the significance of integrating multiple architectural components to enhance the learning capability of neural networks for time-series forecasting tasks, as demonstrated by the hybrid GRU-LSTM model in comparison to LSTM alone.