Indirect measurement of key water quality parameters in sewage treatment plants

Abstract

Many variables are regularly measured in frequently taken water samples from sewage treatment plants. This results in large data sets with associated high costs. The total phosphorus in the sewage is of great importance since it is an indicator of the total pollution concentration, but no direct on-line analyser is available.This paper concentrates on finding an indirect model for total phosphorus content (tot-P) based on other variables routineiy analysed in inlet water to a Nonvegian sewage treatment plant. Data from 1993 were used for calibration and the developed models were validated with data from 1994. A PLS model for tot-P based on eight X-variables gave the lowest test set RMSEP (0.35). Models based on fewer X-variables gave higher RMSEP, indicating that more X-variables stabilize the models. For practica1 use a model for tot-P based on orthophosphate, total organic carbon and suspended solids seems to be a good compromise between simplicity and predictability. With this model the total phosphorus content was predicted with an average deviation (RMSEP) of 0.45 mgP 1 -' from the measured values, giving a relative precision of 20%. Since the X-variables in the model can be measured on-line, this model can be used as an input for coagulant dosing in the sewage treatment plant and for process control. For this purpose the 20% relative precision is quite acceptable. Preliminary results of PLS models for alkalinity, chemical oxygen demand and biologicai oxygen demand indicate that useful modeis can also be developed for these Y-variables.The use of these PLS models has a substantial cost saving potential. They may replace traditional chemicai analysis, speed up the time from sampling to results and represent a possible input to on-line process control. O 1996 by John Wiley & Sons, Ltd.

Knowledge based recursive non-linear partial least squares (RNPLS)

“…The plant can be divided into various stages: a primary clarifier, activated sludge reactors where biological reactions are carried out to remove nitrogen and the organic matter, a secondary clarifier, where clean water is obtained, and an anaerobic digester where pathogenic microbes are removed from the sludge to make them suitable to be sent to a dumping site. A complete review on data driven soft sensors for wastewater treatment plants can be found in [64] and the first applications of PLS in WWTP to predict process variables are in [65,66,67].…”