The article presents the results of the study of the possibilities of using neural networks to solve the problem of determining the active set of a wind farm (WF), taking into account the efficiency of each wind turbines (WT). The comparative analysis of the obtained results with the known methods of determining the active composition of WF, such as: the method of dynamic programming; the method of dynamic programming with increasing the load on the experimentally set percentage; modified method of dynamic programming. The advantages and disadvantages of using each of the studied methods in terms of the possibility of achieving a given generation power at the maximum efficiency of the selected WT are determined. It is established that when using recurrent neural networks to solve the problem of determining the active composition of WF, the minimum direct linear variation of the difference between the power to be generated and the actual power of the determined active set of WF is 2.7%. Under the same conditions, the use of other known methods, in particular, the modified method of dynamic programming ensures the achievement of this parameter at the level of 0.05%. This significantly increases the time to solve the problem. By computer simulation, it was found that under equal conditions, the time to solve the problem using neural networks - 0.04 s, and using a modified method of dynamic programming 3.4 s. The obtained results provide an opportunity to implement effective decision support systems in energy flow management.