TY - JOUR
T1 - A Foresight-Seeing and Transferable Optimization Method for Synergic Operation of Multiple Flexible Resources in Active Distribution Network
AU - Xia, Shiwei
AU - Wang, Yifeng
AU - Li, Haiyang
AU - Li, Gengyin
AU - Zhu, Ziqing
AU - Lu, Xi
AU - Shahidehpour, Mohammad
N1 - This work is supported in part by the National Key R&D Program of China (2021YFB1600205), the National Natural Science Foundation of China (52077075), and the Fundamental Research Funds for the Central Universities (2023JC001).
Publisher Copyright:
© 1972-2012 IEEE.
PY - 2024/9/17
Y1 - 2024/9/17
N2 - With a large number of flexible resources accessing the active distribution network (ADN), the security and economic operation of ADN face more challenges. In this paper, the flexible operation portrait model of electric vehicles (EVs) is first established, and a Bi-directional Long Short-Term Memory (BiLSTM) based method is proposed for predicting the entry and departure information of EVs. Furthermore, a collaborative optimal operation model of multiple flexible resources including soft open points (SOPs), distributed generations (DGs), EVs and dynamic network reconfiguration is proposed for ADN optimal operation. In order to solve the model, the operating states of flexible resources are transformed into the state space, and the double deep Q network (DDQN) solution algorithm is designed to efficiently solve the ADN optimal operation strategy. Moreover, DDQN is enhanced with the transfer learning (TL) mechanism to form a DDQN-TL algorithm, which would well adapt to significant changes in ADN operation environments and avoid the expensive time consumption of retraining of DDQN. Finally, simulation results validated the effectiveness of the proposed ADN optimal operation model and DDQN-TL algorithm for improving ADN operation security and economics.
AB - With a large number of flexible resources accessing the active distribution network (ADN), the security and economic operation of ADN face more challenges. In this paper, the flexible operation portrait model of electric vehicles (EVs) is first established, and a Bi-directional Long Short-Term Memory (BiLSTM) based method is proposed for predicting the entry and departure information of EVs. Furthermore, a collaborative optimal operation model of multiple flexible resources including soft open points (SOPs), distributed generations (DGs), EVs and dynamic network reconfiguration is proposed for ADN optimal operation. In order to solve the model, the operating states of flexible resources are transformed into the state space, and the double deep Q network (DDQN) solution algorithm is designed to efficiently solve the ADN optimal operation strategy. Moreover, DDQN is enhanced with the transfer learning (TL) mechanism to form a DDQN-TL algorithm, which would well adapt to significant changes in ADN operation environments and avoid the expensive time consumption of retraining of DDQN. Finally, simulation results validated the effectiveness of the proposed ADN optimal operation model and DDQN-TL algorithm for improving ADN operation security and economics.
KW - deep reinforcement learning
KW - distributed generation
KW - distribution network optimization
KW - electric vehicle
KW - soft open point
UR - http://www.scopus.com/inward/record.url?scp=85204464188&partnerID=8YFLogxK
UR - https://ieeexplore.ieee.org/document/10682465/authors
U2 - 10.1109/TIA.2024.3462900
DO - 10.1109/TIA.2024.3462900
M3 - Journal article
AN - SCOPUS:85204464188
SN - 0093-9994
JO - IEEE Transactions on Industry Applications
JF - IEEE Transactions on Industry Applications
ER -