Аңдатпа
The growing demand for freshwater, driven by population growth, industrialization, and climate change, has increased global reliance on seawater desalination. While reverse osmosis (RO) remains the primary technology for salt removal, ultrafiltration (UF) plays a critical role as a pretreatment stage by reducing suspended solids, colloids, and microorganisms contributing to RO membrane fouling. This study focuses exclusively on the predictive modelling and real-time optimization of a UF system operating as a pretreatment component in a seawater desalination plant. Using 426 days of operational data, four Machine learning (ML) models: Tree Regression (TR), Ensemble Learning (ENS), Neural Networks (NN), and Gaussian Process Regression (GPR) were trained to predict UF flow rates and membrane resistance in real time. ENS achieved the best performance with an R² of 0.99 and Root Mean Square Error (RMSE) of 3.08 L/min, closely followed by TR (R² = 0.99, RMSE = 3.27 L/min). NN and GPR yielded R² of 0.98 (RMSE = 4.69 L/min) and R² of 0.97 (RMSE = 5.98 L/min), respectively. Adaptive backwash control guided by these predictions reduced average TMP excursions by 18 % and decreased backwash frequency from one event every 6 h to one every 8 h, improving operational stability and cutting maintenance costs by 12 %. This framework presents a scalable, intelligent approach to fouling mitigation in seawater desalination pretreatment. • Effects of time, temperature, pressure, pH, and ORP on ultrafiltration performance were investigated. • Ensemble learning and tree regression predicted flow rate with R² = 0.99 and RMSE = 3.08 L/min. • A 426-day seawater UF dataset enabled robust long-term model training and validation. • Real-time sensors improved fouling detection, backwash timing, and energy efficiency. • The ML framework allowed adaptive control of TMP and flow rate, scalable to other membrane systems.