Chemometric modeling of a membrane ultrafiltration process in a pilot water treatment plant

摘要

A multiple projection to latent structures (PLS) model for predicting short term foulingrnrate (STFR) and backwash (BWR) and air scour (AIRS) cleaning recovery of an ultrafiltrationrnprocess installed in a water treatment pilot plant was developed. The modelsrnwere calibrated and validated with data collected over 3 years of membrane operation.rnThe target variable of the STFR model was the characteristic STFR of the membrane forrneach filtration cycle while the input variables were the filtration resistance at thernbeginning of the cycle, membrane operational and water quality parameters. The resultsrnshowed that this model was able to capture large variations in STFR but at the samerntime it performed poorly in the resolution of local details. On the other hand, it wasrnobserved that the model did not degrade when applied to the validation data set over arntime period of 1 year. This result proves that the model captured the intrinsic nature ofrnthe process and that the apparent low level of variance captured was caused by randomrnperturbations which, the modeling technique was able to filter very efficiently.