A Comparative Study of Reinforcement Learning Algorithms for Distribution Network Reconfiguration with Deep Q-learning-based Action Sampling

Distribution network reconfiguration (DNR) is one of the most important methods to cope with the increasing electricity demand due to the massive integration of electric vehicles. Most existing DNR methods rely on accurate network parameters and lack scalability and optimality. This study uses model...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE access Ročník 11; s. 1
Hlavní autoři: Gholizadeh, Nastaran, Kazemi, Nazli, Musilek, Petr
Médium: Journal Article
Jazyk:angličtina
Vydáno: Piscataway IEEE 01.01.2023
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Algorithms
Comparative studies
data-driven control
deep Q-learning
Distribution network reconfiguration
Distribution networks
Electric power demand
Electric vehicles
Load modeling
Loss reduction
Machine learning
Machine learning algorithms
Mathematical models
Optimization
proximal policy optimization
Q-learning
Reconfiguration
reinforcement learning
Sampling methods
soft actor-critic
Switches
ISSN:2169-3536, 2169-3536
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Distribution network reconfiguration (DNR) is one of the most important methods to cope with the increasing electricity demand due to the massive integration of electric vehicles. Most existing DNR methods rely on accurate network parameters and lack scalability and optimality. This study uses model-free reinforcement learning algorithms for training agents to take the best DNR actions in a given distribution system. Five reinforcement algorithms are applied to the DNR problem in 33- and 136-node test systems and their performances are compared: deep Q-learning, dueling deep Q-learning, deep Q-learning with prioritized experience replay, soft actor-critic, and proximal policy optimization. In addition, a new deep Q-learning-based action sampling method is developed to reduce the size of the action space and optimize the loss reduction in the system. Finally, the developed algorithms are compared against the existing methods in literature.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2023.3243549