Design of an artificial neural network model to predict optimum coagulant dosage in a water treatment plant : case study : Bungokho water works.

Abstract

The prevailing challenges faced by the water industry is the requirement to produce high quality treated water as stipulated by the regulatory authorities and also achieve production requirements at a lower cost. However, ensuring optimized dosing-of the coagulant at conventional water treatment plants, has, the potential achieve both of these objectives. This will result in water quality improvement and also generate chemical cost savings where potential overdosing of coagulant IS minimized. However, optimum coagulant dosage at Bungokho water works is evaluated using a jar test, of which this process is highly time-consuming for operators where results can only be obtained after several hours and does not allow adjustment of alum dose rates to keep pace with rapidly changing raw water quality, Low dosage or under dosing generally results in poor removal of the raw water turbidity, thus failure to meet the water quality targets and less efficient operation of the water treatment plant. Additionally, excessive coagulant or overdosing, leads to-more sludge forming (which are difficult to dewater), chemical wastage and an increase in the operational cost of the treatment This paper addresses the problem of determination of optimal coagulant dosage from raw water characteristics such as turbidity, pH and colour using artificial Neural Networks with the use of ANN, it introduces criteria, given for selection and optimization is done very fast, efficiently before the water is supplied to the public. Artificial Neural Networks have, been preferred for this project for their ability to model nonlinear phenomena recognisable patterns within years of the experimental data and their ability to model nonlinear phenomena. The performance of the ANN model was tested using statistical analyses it and proved to be outstanding with MSE of 0.5008, Coefficient determination of 0.4936 RMSE MSE of 0.4936 and MBE 0f 0.2063. The model was developed and programed using Mat lab 2013 environment.