Modelling of secondary clarifier using regression analysis and artificial neural networks

Mathematical models for the surface area of secondary clarifier were developed for wastewater generated from a dairy industry and from domestic sources, by correlating the parameters namely, surface area per unit flow rate (A/Q), influent concentration (C(O)), underflow concentration (C(U)), recycling ratio (r) and Mean Cell Residence Time (theta C) using multiple regression analysis. There was found a good correlation between the measured data and the model results with regression coefficients of 0.9. Thickener area requirement of combined wastewater was comparehat obtained for dairy wastewater. Thickener area was found to decrease with increase in Mean Cell Residence Time and the area required for treating the combined wastewater was less, when compared with the requirement for dairy wastewater treatment. Neural network was trained with experimental data to 'acquire' knowledge about it. The Back Propagation Network technique was used in which the error was back propagated through the network. The results evolved from the neural network training were compared with the results of regression model and experimental data. Greater deviation was observed between the observed and predicted values of A/Q at high underflow concentrations, indicating that the limiting solids flux was reached. The output from Neural Network approach had greater consistency with the experimental data than the output from conventional regression analysis. Hence, Artificial Neural Network technique is highly adaptive and efficient in investigating input - output relationships.