A multi-agent deep reinforcement learning approach enabled distributed energy management schedule for the coordinate control of multi-energy hub with gas, electricity, and freshwater. (1st March 2022)

• Record Type: Journal Article
• Authors: Zhang, Guozhou; Hu, Weihao; Cao, Di; Zhang, Zhenyuan; Huang, Qi; Chen, Zhe; Blaabjerg, Frede
• Is Part Of: Energy conversion and management. Volume 255(2022)
• Publication Date: 2022-03-01
• Subjects: Multi-energy hub -- Cost reduction -- Distributed energy scheduling policy -- Attention mechanism -- Multi-agent deep reinforcement learning; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2022.115340
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: A wind and photovoltaic driven multi-energy hub model with gas, electricity and freshwater is modelled. The energy management problem of multi-agent system is formed as a partially-observable Markov game. A novel attention mechanism-based multi-agent deep reinforcement learning method is applied. The distributed coordinated control of each energy hub in the system is achieved. Abstract: In recent years, due to the deeply concerns on environment protection, the production, transformation and utilization of the energy sources with a more efficient and various way has become an important research topic. Under this background, this paper designs a renewable energy powered multi-energy hub system with gas, electricity, and freshwater sub-system. To enhance the flexibility and reduce the total cost of such system, the energy management of the multi-energy hub is formed as multi-agent cooperative control, and several targets, including operational costs, environment cost, and maintenance cost are considered along with the constraints. Subsequently, a novel attention mechanism-based multi-agent deep reinforcement learning algorithm is applied, where multi-agents are centrally trained to obtain the coordinate energy management strategy while being executed in a decentralized manner to provide the dispatch instruction for each energy hub with only local states. Finally, the effectiveness of the proposed method is investigated in the study system and the simulation results show that it can reduce the total cost by up to 7.28% compared with the other benchmark methods.