Pattern recognition of fractal profiles in coagulation-flocculation process of wastewater via neural network

Abstract

The optimum chemical dosage is presumably the goal of every chemical water or wastewater treatment plant as in a coagulation-flocculation process. However, due to difficulties in on line measurement and complexity of chemical reaction, the optimum dosage is very hard to be determined by conventional methods. This paper presents a new low cost sensor to measure size changes of flocs that are being formed during a coagulation flocculation process by measuring fractal dimensions. It is based on real time images of flocs that are being captured during the coagulation-flocculation process. A neural network model using Matlab version 6 (The Mathworks Inc., U.S.) was developed in order to recognise the general pattern of fractal profiles in a coagulation-flocculation process of wastewater. Back propagation neural networks (BPN), which are a type of feed-forward networks, with 10 neuron on an input layer, 1 neuron on an output layer, 16 neurons on a hidden layer were used with log sigmoid activation functions. In this BPN model, the gradient descent momentum was used to minimize the errors. The training and testing of such networks were based on input output data from real-time experimental runs on the coagulation-flocculation process. The network converged with 7181 epoch. The developed neural gave good recognition performance with 95% success rate.