Multivariate Predictive Models for Fraser Papers Effluent Treatment Plant at Edmundston, 2000 Environmental Conference Proceedings

- S. Graff, G. Broderick, P. Tessier, R. Leger -- Empirical dynamic models were developed to provide the personnel at the Frasier Papers pulp mill in Edmundston with a short-term prediction of the effluent quality after bioligical treatment. Multivariate statistical analysis of daily operating data (flow rate, pH, temperature, TSS, BOD5, COD, etc...) collected over an 18 month period was used to build Partial Least Square (PLS) dynamic models. Three statistical models are presented and their prediction performances are discussed. The dynamic PLS model implemented in Edmundston predicts the treated effluent total suspended solids (TSS), biological oxygen demand (BOD5)and chemical oxygen demand (COD) from 1 to 4 days in advance based on the untreated effluent past and current characteristics as well as the final effluent history. The developed models explained 67% of the variation of the TSS, 59% for the COD and 39% for the BOD5. These models were implemented at Edmundston in January 1999. During the year, the mill made several process changes to increase lagoon performance including increasing aeration capacity, and increasing retention time by reducing effluent flow and dredging the basin. The predictive final effluent model was able to adapt to these changes and for the last twelve-month period, the TSS model explained 67% of the variation in the final effluent TSS.