Développement d'outils d'aide à l'opération du système de coagulation-floculation-décantation de l'usine de traitement des eaux de Sainte-Foy
|Advisor:||Bouchard, Christian; Grandjean, Bernard|
|Abstract:||The coagulation process is the first step of the conventional drinking water treatment chain. It is an important treatment step since it affects the efficiency of the subsequent treatment steps namely flocculation, settling, filtration and disinfection. It is relevant to develop decision aid tools to assist operators in the choice of the coagulant dose. This project aims at developing such tools. More specifically, the objective of the study was to provide tools for the operators of the Sainte-Foy water treatment plant to help them in choosing the appropriate aluminum sulphate dose (alum). As part of this project, three tools were developed: a model for the prediction of the coagulant dose, two models for the prediction of dissolved organic carbon (DOC) concentration of settled water and a virtual sensor which allows predicting DOC concentration of raw and settled waters. All models are neural network models. The first model allows the prediction of the alum dosage by mimicking the good previous operation performed at the plant in terms of turbidity reduction. The input variables of the model are the month, the conductivity, temperature, turbidity and pH of raw water. The model was developed from operation data collected every 5 minutes during 4 years (378 535 data sets). Dosages predicted vary by an average of 5,9% of those actually applied. The second model allows the prediction of the DOC of the settled water. The input variables are the UV absorbance and DOC of raw water, pH of coagulation and alum dosage applied. Performances of the second model are compared with those obtained from two others empirical models (from the literature) that allow the prediction of the DOC of the settled water. Compared to these models, the second neural model gives better prediction performance. DOC concentrations predicted by the second model vary by an average of 9,6% of those actually measured. The third model allows the prediction of the DOC of raw and settled water. The input variables are the UV absorbance, temperature, turbidity and pH. The model acts as a virtual sensor of DOC concentration and allows the evaluation of the removal efficiency of natural organic matter by the coagulation, flocculation and settling steps. DOC concentrations predicted by the third model vary by an average of 13,2% of those actually measured. Finally, the fourth model allows the prediction of the DOC of settled water from UV absorbance of raw water instead of DOC. Concentrations predicted by that model vary by an average of 10,7% of those actually measured. The database for the adjustment of the second, third, and fourth models includes one year of DOC and UV absorbance monitoring at raw and settled water performed twice daily and operation data continuously collected. The models performances are presented and discussed according to their implementation and use in the treatment plant. A way to improve developed models is also described. Actually, only the first model could be implemented on a short term basis. Models 2, 3 and 4 are actually preliminary models that would need to be updated with larger databases including more variation periods before implementation. Developed models could be integrated to allow the operators to choose the alum dosage that can afford to make a compromise between the different objectives of the coagulation process. This could further improve the treated water quality.|
|Document Type:||Mémoire de maîtrise|
|Open Access Date:||18 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.