The article analyzes the relationship between power generation at hydropower plants (HPPs) and temperature changes for medium-term forecasting in the isolated power system of the Gorno-Badakhshan Autonomous Oblast (GBAO) of the Republic of Tajikistan. Improving the accuracy of forecasting will solve the problem of controlling water consumption, as well as optimize the generation of electricity at the HPP with the provision of reliable functioning of the power system. The solution of such problems is associated with a number of problems such as the lack of sufficient data, the uncertainty of power generation, the lack of regularity of one station operation and poorly reliable forecasting models. In the medium-term forecasting of electricity generation at HPPs, the seasonality of changes in water flow and inflow should be taken into account, especially in power systems with a high proportion of renewable energy sources, where temperature changes directly affect reserves and the possibility of regulation. The paper considers the problem of constructing a model for medium-term forecasting of electricity generation at HPPs taking into account temperature changes in isolated power systems. As a method of medium-term forecasting of power generation, an approach based on machine learning methods was chosen, which is characterized by a high degree of self-adaptation in case of sudden changes in weather conditions. A comparative study of such models as linear/polynomial regression with Tikhonov regularization, k-nearest neighbours, adaptive boosting of decision trees, adaptive boosting of linear models, random forest, extreme gradient boosting, multilayer perceptron. As a result of experimental and industrial calculations, the expediency of using a model based on adaptive boosting with linear regression (ABLR) has been proved.