Coordinated Scheduling Strategy for Multi-Agent System in Active Distribution Grid based on Deep Reinforcement Learning Method

With the increasing complexity of the power system, the traditional centralized scheduling strategy has been unable to meet the demand for efficient management and stable operation of the system, and the multi-agent collaborative regulation technology has gradually become a key means of power system...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2024 3rd International Conference on Energy and Electrical Power Systems (ICEEPS) S. 1410 - 1415
Hauptverfasser: Liu, Xuan, Li, Wanbin, Song, Lin, Yao, Zhanfeng, Lu, Tianguang, Bai, Xue, Zhang, Yuqi, Wang, Wenxin, Zheng, Yanan, Liu, Chunxiu
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 14.07.2024
Schlagworte:
Collaboration
Complexity theory
coordinated scheduling
Deep reinforcement learning
Intelligent agents
Multi-agent optimization operation
Multi-agent systems
power distribution network
Power system management
Real-time systems
Regulation
Resource management
Smart grids
variable renewable energy
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:With the increasing complexity of the power system, the traditional centralized scheduling strategy has been unable to meet the demand for efficient management and stable operation of the system, and the multi-agent collaborative regulation technology has gradually become a key means of power system management. In this paper, a coordinated scheduling strategy for multi-agent system in power distribution network based on deep reinforcement learning method is proposed. First, each subject such as distributed power sources, energy storage devices and flexible loads is taken as an intelligent agent in an active distribution grid, and the real-time state information of different intelligent agents in the grid, including voltage, current, power, etc., is obtained. And then, based on the real-time state information, the expected cumulative rewards of taking load management actions under a given smart grid state are evaluated by a neural network representation of the Q-value function, in which a single intelligent agent obtains the optimal energy allocation strategy by maximizing the Q-value function. This paper also constructs a Maldivian network communication model, based on which information exchange between multiple intelligent agents is realized. Finally, a deep deterministic policy gradient algorithm is used to decide the energy allocation of the multi-agent system, and the effectiveness of the proposed method is verified.
DOI:10.1109/ICEEPS62542.2024.10693015