Title : Data-driven modeling and optimization of the coagulation process in water treatment using RSM and neural networks
Abstract:
Considering the growing demand for drinking water and the increasing use of chemical agents in various purification stages, this study investigated the optimal dosage of PAC (Polyaluminum Chloride) as a coagulant under different conditions. Laboratory-synthesized samples with varying turbidity levels were analyzed using the Response Surface Methodology (RSM). Results showed that increasing the coagulant dose initially enhanced turbidity removal but eventually led to diminishing returns. Moreover, at lower pH levels, removal efficiency was higher, and the final turbidity was lower. Given that the Jar Test is time-consuming and prone to human error, it cannot always be performed accurately or relied upon for consistent results. Therefore, in this study, both Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF) neural networks were employed to predict the outcomes of the Jar Test. Additionally, Principal Component Analysis (PCA) was used to preprocess the data, after which the MLP and RBF models were applied again. Among the two models, the RBF network demonstrated superior predictive performance, evidenced by higher R² values and lower error rates. Furthermore, applying PCA improved the performance of both neural network models.
Keywords: response surface methodology, jar test, neural network, MLP, PCA, RBF
