Abstract:
In this study, the Sidi Khelifa drinking water treatment system
(Oued El Athmania) was selected as the application site. The data
used were collected over a representative period and included: raw
water turbidity, pH, temperature, and conductivity. A deep neural
network (DNN) model was developed and trained in MATLAB
using the Levenberg-Marquardt learning algorithm. Several
architectures were tested to identify the optimal network structure.
The activation functions used included the sigmoid hyperbolic
tangent for the hidden layers and a linear function for the output.
The results obtained showed that the DNN model is capable of
accurately predicting the optimal dosage of coagulant (12 mg/L),
with a correlation coefficient greater than 0.95 between the
predicted and measured values during the training, validation, and
testing phases. The model thus makes it possible to anticipate
reagent needs, adjust dosages in real time, and ensure that the
treated water complies with drinking water standards. This
intelligent modeling approach can prove particularly useful in the
field of drinking water treatment, where raw water quality is
subject to significant seasonal fluctuations. It helps reduce
overdosing, improve treatment efficiency, and control costs while
ensuring better protection of public health.
