TY - JOUR
T1 - Deep neural network-based hierarchical learning method for dispatch control of multi-regional power grid
AU - Tang, Hao
AU - Lv, Kai
AU - Bak-Jensen, Birgitte
AU - Pillai, Jayakrishnan Radhakrishna
AU - Wang, Zhengfeng
N1 - Funding Information:
The research is supported by State Grid Corporation of China Project “Intelligent Scheduling Technology based on Deep Learning in Flexible Environment” (SGTYHT/19-JS-215).
Publisher Copyright:
© 2021, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.
PY - 2021
Y1 - 2021
N2 - Multi-regional power grid with interconnected tie-lines has become an increasingly important structure for current power systems, and can efficiently reallocate power resources on a large scale. The power dispatch of a multi-regional power grid involving multiple resources plays a key role in maintaining system balance and improving operating profit. Current optimisation methods for this dispatch problem need to execute a complete optimisation calculation at each dispatch moment, and lack online decision and optimisation abilities. Therefore, we introduce a deep neural network-based hierarchical learning optimisation method to establish an online approach to focused coordination dispatch problems. The method can realise system optimisation based solely on historical operating data. First, the focused coordination dispatch problem is formulated mathematically. Then, we establish a hierarchical structure suitable for online learning methods. Under this designed structure, we establish a learning optimisation model for each agent, and introduce a deep reinforcement learning algorithm for solving the optimisation problems online. Simulation results based on the IEEE 300-bus system are presented to validate the efficiency and availability of the proposed hierarchical method.
AB - Multi-regional power grid with interconnected tie-lines has become an increasingly important structure for current power systems, and can efficiently reallocate power resources on a large scale. The power dispatch of a multi-regional power grid involving multiple resources plays a key role in maintaining system balance and improving operating profit. Current optimisation methods for this dispatch problem need to execute a complete optimisation calculation at each dispatch moment, and lack online decision and optimisation abilities. Therefore, we introduce a deep neural network-based hierarchical learning optimisation method to establish an online approach to focused coordination dispatch problems. The method can realise system optimisation based solely on historical operating data. First, the focused coordination dispatch problem is formulated mathematically. Then, we establish a hierarchical structure suitable for online learning methods. Under this designed structure, we establish a learning optimisation model for each agent, and introduce a deep reinforcement learning algorithm for solving the optimisation problems online. Simulation results based on the IEEE 300-bus system are presented to validate the efficiency and availability of the proposed hierarchical method.
KW - Coordination dispatch
KW - Deep reinforcement learning
KW - Hierarchical optimisation
KW - Multi-regional power grid
UR - http://www.scopus.com/inward/record.url?scp=85104827567&partnerID=8YFLogxK
U2 - 10.1007/s00521-021-06008-4
DO - 10.1007/s00521-021-06008-4
M3 - Review article
AN - SCOPUS:85104827567
SN - 0941-0643
JO - Neural Computing and Applications
JF - Neural Computing and Applications
ER -