Modelling the operation of a Water Treatment Plant based on Artificial Neural Networks

The main purpose of this study is to model the operation of a Drinking Water Treatment Plant (DWTP) using its main operational and water quality parameters in a fast, easy and reliable way. This study is based on a large number of data from recent years (2019-2021). The DWTP has a maximum capacity of 110,600 m3/day and is located at Hersonissos, Crete in Greece. The methodology that was followed comprised of the development of Artificial Neural Networks (ANN) in the MATLAB programming environment. Since the 1990s the ANN modelling approach has gained popularity for prediction and forecasting due to its ability to capture complex nonlinear relationships. Two models were developed with satisfactory results with regards to Mean-Square Error (MSE) and Regression Coefficient (R) values. The models were able to predict the main operational parameters such as the dosages of coagulants, flocculants and disinfection (O3, Cl2(g)) chemicals rendering them a useful tool for the DWTP operator. For future work a greater number of tests are planned to check different ANN input parameters and architectures with different numbers of hidden neurons.