Affiliations: [a] CIRSEE – Suez Lyonnaise des Eaux, Technopolis, ZAC de Mercières, 14, rue du Fonds Pernant, F-60471 Compiègne, France. E-mail: [email protected] | [b] Université de Technologie de Compiègne, UMR CNRS 6599 Heudiasyc, BP 20529, F-60205 Compiègne, France. E-mail: [email protected]
Abstract: This paper reports on the application of Artificial Neural Network techniques to coagulation control in drinking water treatment plants. The coagulation process involves many complex physical and chemical phenomena which are difficult to model using traditional methods. The amount of coagulant ensuring optimal treatment efficiency has been shown experimentally to be non-linearly correlated to raw water characteristics such as turbidity, conductivity, pH, temperature, etc. The software sensor developed is a hybrid system including a self-organising map (SOM) for sensor data validation and missing data reconstruction, and a multi-layer perceptron (MLP) for modelling the coagulation process. A key feature of the system is its ability to take into account various sources of uncertainty, such as atypical input data, measurement errors and limited information content of the training set. Experimental results with real data are presented.
Keywords: data validation, missing data reconstruction, outlier rejection, self-organising maps, multi-layer perceptrons, hybrid system, coagulation control, water treatment
