:   .., .., ..
:  
:  121
:  
:  2026
:   .., .., .. // . - 2026. - . 121. - .127-143.
:  , , Q- , , ,
(.):  microgrids, deep reinforcement learning, Q-function of utility, isolated power system, active and reactive node power, agent training
:   . , , , . , , . . . . , . , , .
(.):  The process of designing adaptive distributed electric power systems using deep reinforcement learning is considered. Adaptive systems are microgrids, which are local, autonomous distributed electric power systems with different topologies depending on the state of the electrical network. For the correct formation of microgrids, topological and operational constraints have been identified, on the basis of which adeep reinforcement learning agent determines possible actions that bring it closer to achieving a given goal. The task is to maximize the total capacity of the powered nodes in the network, taking into account their priorities and the specified limitations. An algorithm for solving this problem through reinforcement learning is presented. Anumerical example demonstrates the process of forming microgrids for a test network. The high proportion of powered network power due to the actions of deep reinforcement learning agents confirms the assumption that this approach is correct and justified for the task under consideration. A computational experiment has shown that with a changing topology of the source network, it is possible to provide electricity in real time using the same solution method that dynamically adapts to different conditions.

PDF

: 98, : 33, : 5.


© 2007.