A multi-agent decision approach for optimal energy allocation in microgrid system

In this article, we investigate the energy allocation problem for microgrids with energy supply coordination. To meet the requirements of the manufacturing system and reduce the operation cost, we propose a microgrid systems control model based on multi-actor-attention-critic with the curriculum learning (CL-MAAC). More specifically, the allocation problem of energy is fabricated as a Markov decision process (MDP). Meanwhile, all energy supplies are abstracted as multiple independent agents. Then, a multi-actor-attention -critic reinforcement learning (RL) framework is employed for encouraging each agent to coordinate with other agents on the basis of different attention. Besides, a curriculum learning (CL)-based sampling policy is adopted to select the transitions that are more favorable to the model based on the degree of difficulty. This integration method of CL and multi-actor-attention-critic significantly enhance the robustness and effectiveness of the algorithm. A simulation of the energy allocation in the microgrid system has been implemented, which demonstrates the great performance of CL-MAAC.