Using Multilayer Perceptron Artificial Neural Network for Predicting and Modeling the Chemical Oxygen Demand of the Gamasiab River

• Main Authors: Mohamad Parsimehr (Author), Kamran Shayesteh (Author), Kazem Godini (Author), Maryam Bayat Varkeshi (Author)
• Format: Book
• Published: Hamadan University of Medical Sciences, 2018-06-01T00:00:00Z.
• Subjects: article
• Online Access: Connect to this object online.
• Notes: 2423-4583; 10.15171/ajehe.2018.03
• Keywords: Artificial Intelligence; River Quality; Environmental Assessment; Environmental sciences; GE1-350; Public aspects of medicine; RA1-1270
• Source: Avicenna Journal of Environmental Health Engineering, Vol 5, Iss 1, Pp 15-20 (2018)

Concerns about water quality have widely increased in the last three decades; thus, water quality is now as important as its quantity. To study and model the quality of the Gamasiab River, its data, including chemical oxygen demand (COD), biological oxygen demand (BOD), dissolved oxygen (DO), total dissolved solids (TDS), total suspended solids in water, acidity, temperature, turbidity, and cations and anions were measured at four stations. Then, the correlations between these parameters and COD were measured using Pearson's correlation coefficient and modeled by multilayer perceptron artificial neural network. In order to minimize the cost of the experiments performed and to provide the input parameters to the artificial neural network based on the correlations between the data and COD, the number of input parameters was reduced and finally, model No.3, with the Momentum training function and the TanhAxon activation function with the validation correlation coefficient of 0.97, mean absolute error of 2.88, and normalized root mean square error of 0.11 was identified as the most accurate model with the lowest cost. The results of the present study showed that the multilayer perceptron neural network has high ability in modeling the COD of the river, and those data correlated with each other have the greatest effect on the model. Moreover, the number of input parameters can be reduced in order to lower the cost of experiments while the performance of the model is not undermined.