Medium-Term Forecast of Water Inflow to the Sayano-Shushenskaya Hydroelectric Power Plant Reservoir by Autoregressive Machine Learning Models : доклад, тезисы доклада
доклад, тезисы доклада, статья из сборника материалов конференций
Библиографическое описание:Medium-Term Forecast of Water Inflow to the Sayano-Shushenskaya Hydroelectric Power Plant Reservoir by Autoregressive Machine Learning Models : доклад, тезисы доклада / Natalia F. Cheganova, Pavel V. Matrenin, Sergey V. Mitrofanov. - [S. l. : s. n.], 2023. - Текст : непосредственный // 2023 IEEE XVI International Scientific and Technical Conference: proceedings : PROCEEDINGS APEIE - 2023 : Institute of Electrical and Electronics Engineers, 2023. - P. 470-473. - The reported study was supported by Russian Science Foundation, research project No. 22-79-00181. - ISBN 9798350330885, DOI 10.1109/APEIE59731.2023.10347751.
Аннотация:Hydropower is an important economic sector. It provides for the generation of green power and proves key in power-frequency control in the energy system. Its peculiarity is its strong dependence on hydrological conditions. Therefore, the task of forecasting and planning the operation of hydroelectric power plants in conditions of changing river flow is relevant. Accurate forecasting of the inflow is used to make the most rational regulation of the reservoir, therefore, increases economic efficiency. A study on forecasting the water inflow into the reservoir of the Sayano- Shushenskaya hydroelectric power plant to improve the efficiency of water-power regimes was performed. The machine learning approach was chosen as a method of medium-term inflow forecasting. Such models as linear regression, polynomial regression and the k-nearest neighbor model are compared in the paper. The purpose of study is to determine the depth of retrospective data and the sampling step based on autoregressive machine learning models. The most efficient result on the test sample was obtained for knearest neighbor model with a depth of retrospective data of 24 months and sample step of 1 months. The coefficient of determination was 0,749.
