Prediction of Total Maximum Daily Loads (TMDLs) of pollutants in river by using artificial neural network (ANN) / Khairunnisa Khairudin, Mohamed Syazwan Osman and Syahrul Fithry Senin

• Päätekijät: Khairudin, Khairunnisa (Tekijä), Osman, Mohamed Syazwan (Tekijä), Senin, Syahrul Fithry (Tekijä)
• Aineistotyyppi: Kirja
• Julkaistu: 2020.
• Aiheet: Conference or Workshop Item; PeerReviewed
• Linkit: Link Metadata
• Notes: https://ir.uitm.edu.my/id/eprint/82439/1/82439.pdf
• Keywords: Quantitative analysis; Analytical chemistry

Total Maximum Daily Load (TMDL) studies are crucial in determining a pollutant reduction target and allocates load reductions necessary to the source(s) of the pollutant. Existing modelling approaches to simulate TMDL allocations of point source and non-point source pollutants typically consist of linking watershed model, receiving water transport model, and receiving water quality model. Such deterministic model requires extensive data of the underlying process compared to artificial neural network (ANN) that simulates data based on data-driven method. In this study, biochemical oxygen demand (BOD), chemical oxygen demand (COD), suspended solids (SS), and ammoniacal nitrogen (NH3-N) loads for Muda River is predicted using ANN. The model is developed based on historical monthly concentration data and discharge data from 2013 to 2018 provided by Department of Environment (DOE), Malaysia. These parameters were introduced as inputs, whereas TMDL as outputs of the threelayer feed-forward back-propagation ANN. The learning algorithm used is Bayesian Regularization with tansig transfer function at the hidden layer and purelin transfer function at the output layer. Here, the number of neurons tested to obtain the optimum number of hidden layer nodes is 5, 7, 9, 11, and 13, which run at different epochs: 1000, 2000, and 3000. Model performance was evaluated using mean absolute percent error (MAPE), coefficient of determination (R2), root mean square error (RMSE), and model efficiency (E). The best model for TMDL of BOD is 6:13:1 at epoch 2000 with 0.0004% (MAPE), 1.0 (R2), 0.0005 (RMSE), and 1.0 (E). Meanwhile, the best model for TMDL of COD is 6:5:1 at epoch 3000 with 0.00004% (MAPE), 1.0 (R2), 0.0004 (RMSE), and 1.0 (E). Furthermore, the best model for TMDL of SS is 6:5:1 at epoch 3000 with 0.0038% (MAPE), 0.99 (R2), 0.1 (RMSE) and 1.0 (E). Finally, the best model for TMDL of NH3-N is 6:5:1 at epoch number 3000 with 0.0001% (MAPE), 1.0 (R2), 9.47x10-6 (RMSE) and 1.0 (E). It can be concluded that ANN is an excellent modelling approach to substitute deterministic models for TMDL prediction.