Day-ahead scheduling based on reinforcement learning with hybrid action space

Driven by the improvement of smart grid, active distribution network (ADN) has attracted much attention due to its characteristic management. By making full use electricity price signals for optimal scheduling, total cost ADN can be reduced. However, day-ahead scheduling problem is challenging since future unknown. Moreover, in ADN, some schedulable variables are continuous while discrete, which increases difficulty determining scheme. In this paper, formulated as a Markov decision process (MDP) with continuous-discrete hybrid action space. Then, an algorithm based on multi-agent reinforcement learning (HRL) proposed obtain The adopts structure centralized training and decentralized execution, different methods applied determine selection policy discrete variables. simulation experiment results demonstrate effectiveness algorithm.