Reliable operation of power systems with a significant share of hydropower plants (HPPs) in the energy mix depends in many respects on how accurately the water inflow is forecasted. Taken together, the water inflow prediction and optimal planning of production define energy security, ensure the possibility of protection from floods, and eliminate idle discharges at hydroelectric power plants. The solution of such problems is complicated by lack of reliable information about the water inflow, its being stochastic in nature, a variable electricity consumption pattern, and inaccurate prediction and planning models. Improvement of prediction accuracy is focused on determining the water inventory for planning prospective electricity generation at HPPs taking into account regulation in the medium term. Such regulation is necessary to meet the power system load in the load curve peak and semi-peak parts. The paper considers the problem of constructing a medium-term water inflow prediction model for planning electricity generation for a week ahead taking into account climate changes in isolated operating power systems taking as an example the electric power systems of the Gorno-Badakhshan autonomous oblast in Tajikistan. For taking into account constant climate changes, it is proposed to use an approach based on machine learning methods, which features a self-adaptation capability. Based on the results of accomplished experimental and industry-grade numerical analyses, the expediency of using a model based on an ensemble of regression decision trees has been shown.