Optimization of Pre-Treatment Operation of Industrial Water Using Artificial Neural Network and Genetic Algorithm

Coagulation process is of vital importance for achieving good performance in water pre-treatment units. It usually implements optimum operating conditions that results in highest turbidity removal. The choice, dosage, pH and rate of mixing of coagulant and coagulant-aid are the variables that define optimum operating conditions. In this research, coagulation process of Fajr petrochemical company has been studied. Several jar tests are conducted to determine performance of Alum, Ferric chloride and Poly aluminum chloride as coagulants and anionic poly electrolyte and wheat starch as coagulant-aids at different pH and mixing rates.Optimum operating conditions data obtained in jar tests are used in developing a Neural Network model. This model allows operators to have an estimation of the operating conditions. The success of water pre-treatment depends on fixed feed water quality without frequent need to run jar tests. The predicted result of this network for operating parameters, presented the high consistency with operating data of industrial unit. The maximum relative error for prediction of turbidity is 0.6 % and for total hardness is 2.3 %. The number of neurons of Neural Network hidden layer is optimized and the developed model is validated and tested using a fraction of data that was not utilized in network training. The estimated operating condition for a given feed water quality is implemented in practice and the result of pre-treated water quality matched the expected quality.