Early warning system in drinking water purification plants for modelling both residual turbidity and aluminium

Environmental decision supporting system (EDSS) was developed in this article using the full-scale purification plant's data. After reprocessing the data by K-means clustering, the clusters' results were used to make the three sets of data, for the models; ANN, GA-ANN and GEP to predict residual turbidity, the results were compared. The hybrid genetic-neural model was the best; it gave a smaller network and reduced overfitting. It gave a very good results, RMSE = 0.20 NTU, R = 0.95. In the regard of predicting residual aluminium, a network with 17 neuron in the hidden layer was obtained and gave a result of RMSE = 0.021 mg/L, R = 0.93. At the end, a graphical user interface was generated by MATLAB software. Based on the best networks gained, in order to make the networks more globalised and easier to be used from different kind of users.