Multi-agent Deep Reinforcement Learning Algorithm for Distributed Economic Dispatch in Smart Grid

With the development of large-scale power grids, the issue of distributed economic dispatch has received considerable critical attention. However, due to the existence of some effects such as valve-point effects, the nonconvex objective function remains a major challenge for the distributed optimization problem. This paper proposes a cooperative deep reinforcement learning algorithm for distributed economic dispatch with the nonconvex objective function. In the distributed algorithm, all nodes obtain the value of actions by observing the environment and update state-action-value function in coordination with local neighbors. The state-action-value function is approximated by a neural network, which allows the algorithm to be used for large and continuous state spaces. The advantages of the algorithm are demonstrated through several case studies.