DBPapers
DOI: 10.5593/sgem2017H/33/S12.007

ARTIFICIAL NEURON NETWORK MODELING AS A PREDICTION MODELS OF WATER QUALITY IN THE LAKES SUBJECTED TO THE TECHNICAL TREATMENTS

R. Augustyniak, M. Neugebauer, J. Grochowska, P. Solowiej
Thursday 23 November 2017 by Libadmin2017

References: 17th International Multidisciplinary Scientific GeoConference SGEM 2017, www.sgemviennagreen.org, SGEM2017 Vienna GREEN Conference Proceedings, ISBN 978-619-7408-27-0 / ISSN 1314-2704, 27 - 29 November, 2017, Vol. 17, Issue 33, 51-58 pp; DOI: 10.5593/sgem2017H/33/S12.007

ABSTRACT

Lake water quality determines the possibility of its using for technical purposes (as a source of drinking or industrial water) and for recreational using (swimming, angling, water sports). In modern times the water quality state is deteriorating because of excessive anthropopressure. The attempts of water quality improvement are taken via technical methods implementation. In order to predict the effects of potential restoration actions the great need exists of the models development, which can help to select the optimal restoration method.
The artificial neural network were used for modeling that process, because of fact, that it is implemented for complex non-linear processes. The neural model is the model based on “the black-box”, that means that we create the models basing on the input and output data. During the process of neuron network learning the relations between input and output are calculated as the force of the particular connections between neurons. Created artificial neuron networks have the ability of prediction of the output data for other input data.
In the created model the contents of nutrient components deciding of the water bodies productivity (nitrogen and phosphorus) in the different lake ecosystem components and the data concerning the environmental conditions (e.g. temperature, dissolved oxygen) in the lake before, during and after restoration.
The several kinds of network were tested (MLP, RBF) with different topology (the different number of neurons in the hidden layers). The time period used in modeling was one year long (with one month step), because of periodicity of the processes, which have place in the temperate zone lakes. Five the best models were used for prediction (with the lowest learning error).
The created model enabled the effectiveness assessment of particular restoration method and as a result – the choice of the most optimal one, viewing from the most expected water quality parameters.

Keywords: water quality, neuron modeling, artificial neural network, lake restoration