Regression Models for Forecasting Water Inflow Based on Retrospective Inflow Data and Meteorological Data

Accurate water inflow forecasting is used to make the most rational realize of reservoir, therefore, increases economic efficiency. It is necessary to maximize the use of water resources and installed capacity of hydroelectric units with optimal participation of the hydroelectric power plant in the balance of generation and consumption of the energy system. Planning the hydroelectric power plants operation is impossible without water inflow forecasting. The uneven river flow and the high dependence on natural phenomena make this task difficult, especially due to climate change. It requires processing a large amount of data from long-term observations (hydrological, meteorological, etc.). With the development of computing technologies, machine learning methods can be used for this task. In the current work, the problem of water inflow forecasting to hydroelectric power plant reservoir is considered. Basic machine learning models such as linear regression, k-nearest neighbor and polynomial regression were used. It is considered how and in what amount weather data affect the forecast result. According to the highest coefficient of determination, the best model with the appropriate depth of retrospective and sampling step was selected.