PREDICTION OF GROUNDWATER FLOW TO THE SOUTH BUG RIVER USING ARTIFICIAL NEURAL NETWORKS AND REGRESSION EQUATIONS

Abstract

The most reliable forecasts can be obtained for hydrogeological objects that have signs of determinism in regime changes. First of all, these include systems with an undisturbed mode. However in the network of state hydrogeological monitoring of groundwater, there are no objects with an undisturbed regime due to the spread of direct anthropogenic influence and indirect - due to the changes in climatic conditions caused by it. Significant variability and unpredictability of changes in traditional regime-forming factors (air temperature and precipitation) over the last decades proves the low efficiency of forecasting the level and flow of groundwater using empirical regression equations built on data from the 1980s - early 2000s. It was possible to obtain more reliable results with the help of neural network modelling, which involves working with significant series of contradictory data that change according to an unknown algorithm. The forecast was made for the specific underground flow to the South Bug River in the area of Khmilnyk. The advantages of simulated forecasting over time series forecasting are shown.