Forecasting Long-Term Inflow to a Hydropower Plant Based on Machine Learning Methods

This article considers the problem of forecasting the inflow to a hydropower plant on an annual interval with a discretization of one month. The first part of the article provides an overview of publications on this topic, and also describes the object of study - the Kolyma hydropower power plant and the characteristic features of the hydrograph of the Kolyma River on which it is located. The paper presents an analysis of the correlation between various meteorological factors and water flow at the hydropower plant site. In the final part, based on the selected factors with the highest degree of correlation, forecasting models are developed based on regression and heuristic machine learning methods. Their accuracy was assessed based on such metrics as the coefficient of determination and standard deviation. Inflow forecasts were calculated based on the most accurate models. The results obtained were used to analyze the characteristic features of runoff variability over long-term forecast periods.