On application of artificial neural networks for modeling of water consumption.

Abstract

The article shows the capabilities of artificial neural networks built on radial basis functions for the study of water consumption by various branches of the Don River basin water system. The use of mathematical models in the form of a system of differentiated equations is hampered by the uncertainty of the coefficients in their right-hand sides, which describe the intensities of processes of different natures: precipitation, water consumption by various sectors of the water management complex, water runoff during snow melting, transpiration, infiltration, etc. As a rule, these parameters are random, and the mathematical models describing the water balance are stochastic. The use of neural networks is very fruitful here. Without going into the physical essence of the processes, they can be used to approximate and make reliable predictions, which is a prerequisite for the development of dynamic-stochastic concepts in the management of water resources.