Scholarly record

SIMULATION OF ARTIFICIAL NEURAL NETWORKS FOR ASSESSING THE ECOLOGICAL STATE OF SURFACE WATER

Mikołaj Karpiński

Abstract

An important problem in assessing water quality is that the data on water pollution and their analysis are formed over the previous period, and as a result, it is difficult to assess the current status of water resources, the more likely it is to predict it for the future period, which would allow timely action on improvement of the ecological state of water. The purpose of the work is to evaluate the possibility of using neural networks to simulation of surface water pollution. The data are derived from the results of the monitoring of the environment of Lviv region during 2008-2016 years. The efficiency of the proposed model was determined using the input parameters of concentrations of pollutants (nitrates, suspended matter, dry residue, sulfates) and the output parameter ? the concentration of nitrates in water. The performance of the model was compared to the parameters ? Mean Squared Error and Correlation Coefficient. The optimal architecture of the neural network model was determined after multiple training with the change in the number of neurons in the intermediate layer and setting the minimum value of the median error and the maximum value of the correlation coefficient. In this work, different numbers of neurons from 10 to 19 in each layer were tested. Each topology was repeated three times to avoid accidental correlation due to the random initialization of weight coefficients. The model of multilayer perceptron with 4 input parameters (concentration of nitrates, suspended solids, dry residue and sulfates in water), 15 neurons and 1 output parameter ? concentration of nitrates in water, trained using the Levenberg-Markar algorithm, was proposed. The data obtained from the simulation of artificial neural network data of the correlation coefficients are within 99%, which demonstrates the high accuracy of the model of the neural network. The proposed model can be used to predict changes in the concentration of pollutants in water.

Publication Impact Profile

• Citations
• CrossRef - Citation Indexes: 7
• Scopus - Citation Indexes: 17

• Captures
• Mendeley - Readers: 2

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2018/2.1/s07.088

Publication details

Back to publication list