Management of the hydropower plants requires the economically efficient use of water resources based on the forecasts and simulation models of the hydropower plant and the reservoir. There are various data sources for the water inflow forecasting: meteorological and hydrological posts, Earth remote sensing. However, the problem arises of combining the specified heterogeneous data for aggregated processing with the use of machine learning methods. The research goal is to design an architecture of a system for collecting and processing the data from various sources to operational forecast of the water inflow and the reservoir water-level. It was achieved by analyzing and selecting the sources
and methods for the use of Earth remote sensing data; observing the main principles of hydrological
modeling; assessing the availability of the different data; analyzing the ways of increasing the observability of the hydrological objects by installing additional meteorological and hydrological posts; and designing a technology for the automatic data collection and processing. The most significant results are
developed architecture of the data collection and processing system and the technology for aggregating heterogeneous data with the use of machine learning methods. It is aimed to reduce the error of shortterm forecasting of the water inflow to the reservoir. The significance of the results lies in the fact that the proposed technology was offered and justified for a real hydropower plant; and it can improve the water resources management efficiency: increase the energy generation, minimize the sterile spills, increase the flood forecasting horizon and reduce the risk of flooding during the spring high water.