Development of artificial nueral network model for the analysis of wastewater treatment
A statistical modeling tool called artificial neural network (ANN) is used in this work to predict the performance of wastewater treatment plant (WWTP). Extensive influent and effluent parameters database containing measured data spanning over two years of period was used to develop and train ANN us...
| Main Authors: | , |
|---|---|
| Format: | Monograph |
| Language: | English English |
| Published: |
[s.n]
2012
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/31218/ http://irep.iium.edu.my/31218/1/Full_Report.pdf http://irep.iium.edu.my/31218/2/EndofProjectReport_14_03_12.pdf |
| Summary: | A statistical modeling tool called artificial neural network (ANN) is used in this work to predict the performance of wastewater treatment plant (WWTP). Extensive influent and effluent parameters database containing measured data spanning over two years of period was used to develop and train ANN using ANN toolbox in commercially available software, MATLAB. The data were obtained from one of Sewage Treatment Plant in Malaysia. The input parameters for the ANN were BOD, SS, and COD of the influent, while the output parameters were combination of the effluent characteristics. The networks for single input-single output were compared with those of single input-multiple output. The ANN was developed for raw and screened data and the results were compared for both networks. It was found that the use of data screening is essential to come up with a better ANNs model. From the regression analysis, networks with one hidden layer and 20 neurons were found to be the best one for single input-single output approach. While the best network for the multiple inputs-single output approach was with BOD as outputs and 30 neurons. The second approach which showed a lower RMSE and higher R values was selected. The results show that hybrid (PCA+BODinf) model outperformed its corresponding normal BODinf net and recorded a higher correlation coefficients (R) values for training (0.7362), testing (0.7678) and verification (0.7699) datasets with their respective mean absolute errors (MAE) of 13.75,11.29 and 12.76. |
|---|