An artificial neural network model and design equations for BOD and COD removal prediction in horizontal subsurface flow constructed wetlands

Abstract

A model is presented, which can be used in the design of horizontal subsurface flow (HSF) constructed wetlands. This model was developed based on experimental data from five pilot-scale CW units, used in conjunction with artificial neural networks. The CWs were operated for a two-year period under four different hydraulic residence times (HRT). For the proper selection of the parameters entering the neural network, a principal component analysis (PCA) was performed first. From the PCA and model results, it occurs that the main parameters affecting BOD removal are porous media porosity, wastewater temperature and hydraulic residence time, and a set of other parameters which include the meteorological ones. Two artificial neural networks (ANNs) were examined: the first included only the three main parameters selected from the PCA, and the second included, in addition, the meteorological parameters. The first ANN predicted BOD removal rather satisfactorily and the second one examined resulted in even better predictions. From the predictions of the ANNs, a hyperbolic design equation, which combines zero and first order kinetics, was produced to predict BOD removal. The results of the ANNs and of the model design equation were compared to available data from the literature, and showed a rather satisfactory correlation. COD removal was found to be strongly correlated to BOD removal. An equation for COD removal prediction was also produced.