Artificial Neural Networks to Predict Wastewater Treatment in Different Media Hssf Constructed Wetlands

Artificial neural networks, ANNs were developed to model the biological oxygen demand (BOD), biochemical oxygen demand (COD), and total suspended solids (TSS) treatment in 3 horizontal subsurface flow (HSSF) constructed wetlands (CWs) located in Samaha village, Nile Delta, Egypt. Gravel, hollow plastic bits and shredded tires pieces were used as different wetland treatment media. Three hundred data sets were used, of which 240 and 60 for ANNs calibration and validation respectively. Input variables for the models are influent concentration (Ci), loading rate (q), media surface area (As), and actual velocity (v). Performance of the models in calibration and validation processes was evaluated using the error and the percentage error between experimental and model values. Output result is the effluent concentration (Co). Training procedure for effluent concentrations was quite successful; as perfect matches were obtained between measured and calculated concentrations. The ANN that showed best BOD modeling has a structure of 4-5-4-1 (4 input variables, 5 neurons in the first hidden layer, 4 neurons in the second hidden layer, and 1 output variable). As for COD and TSS simulation; the ANN structures were 4-7-5-1 and 4-65-1 respectively. Modeling results displayed very good behavior of the ANN with reliable and accurate simulation. Plastic media gave the best treatment performance than both gravel and rubber media by percentages varied between 6.75 and 10.84% (more than gravel) and between 10.87 and 13.95% (more than rubber) which coped with the measured field results.