This thesis presents two successful applications of data-driven models developed for a reverse osmosis desalination pilot plant. The Support Vector Regression (SVR) modeling approach for steady state operation of the reverse osmosis pilot plant M3 has performed better results than the commonly used first principle-based models. In the same way, the dynamic models have enabled the short-term prediction and simulation of the M3 plant behavior under non-steady state operation, with such a high accuracy that it makes the approach suitable for advanced reverse osmosis plant control algorithms, fault tolerant control and process optimization. In fact, SVR models have been proved to perform excellent results as part of a fault detection and isolation tool for the M3 plant. In addition, an alternative to SVR models which uses Self-Organizing Maps has been also demonstrated for fault detection and isolation, including a useful visual tool for the rapid fault detection during plant operation.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados