Multi-timescale voltage optimization and control for distribution networks based on regional priority evaluation and deep reinforcement learning
Abstract
With the increasing penetration of distributed photovoltaic (PV) generation, energy storage systems, and other flexible resources in distribution networks, conventional voltage regulation methods struggle to address both overvoltage and undervoltage violations and to coordinate heterogeneous devices operating at different timescales. Existing region-based multi-agent control methods typically treat partitioned regions as homogeneous decision-making units and lack an explicit mechanism to characterize spatiotemporal variations in regional regulation capability and available flexibility. To address these limitations, this paper proposes a multi-timescale voltage optimization and control method for distribution networks based on regional priority evaluation and deep reinforcement learning. First, a coordinated voltage regulation architecture spanning long and short timescales is established. At the long timescale, a double deep Q-network (DDQN) is employed to centrally schedule discrete voltage regulation devices from a system-wide perspective. At the short timescale, a comprehensive regional voltage regulation evaluation index is introduced to quantify the regulation capability and potential of each region and thereby determine its regulation priority. A priority-aware state feature encoding (PASFE) module equipped with a learnable gating mechanism then adaptively fuses regional physical information with coordination information broadcast by the upper-level controller. The resulting priority-aware representations are incorporated into the regional actor and centralized critic networks, enabling the multi-agent algorithm to perform fine-grained regulation of individual continuously adjustable devices. Case studies on the IEEE 33-bus and 141-bus systems demonstrate that the proposed method effectively eliminates bus-voltage violations and outperforms the benchmark strategies in mitigating voltage deviations and reducing network losses, thereby confirming the effectiveness of regional-priority guidance in enhancing coordinated multi-timescale voltage regulation.
Related
- Hierarchical dual-timescale deep reinforcement learning for mechanically rotatable STAR-RIS assisted mobile edge computing networks
- Safe reinforcement learning for aggregated demand response control of emerging flexible resources in active distribution networks
- Deep reinforcement learning-driven cross-layer optimization of IRS-assisted NOMA with blockchain for secure 6G oriented networks spectrum management
- Bayesian Neural Networks via Dirichlet Process Priors on the Data-Generating Distribution for Randomized Value Functions in Deep Reinforcement Learning
- Distribution-Agnostic Robust Trajectory Optimization via Chance-Constrained Reinforcement Learning
BibTeX
@article{zhao2026multitimescale,
title = {{Multi-timescale voltage optimization and control for distribution networks based on regional priority evaluation and deep reinforcement learning}},
author = {Shanglin Zhao and Hongsheng Jiang and Teng Xiong and Xiaobo Dou and Qiangsheng Bu},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-75917-y},
url = {https://www.nature.com/articles/s41598-026-75917-y},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multi-timescale voltage optimization and control for distribution networks based on regional priority evaluation and deep reinforcement learning
AU - Shanglin Zhao
AU - Hongsheng Jiang
AU - Teng Xiong
AU - Xiaobo Dou
AU - Qiangsheng Bu
PY - 2026
DA - 2026/10/17
JO - Scientific Reports
DO - 10.1038/s41598-026-75917-y
UR - https://www.nature.com/articles/s41598-026-75917-y
AB - With the increasing penetration of distributed photovoltaic (PV) generation, energy storage systems, and other flexible resources in distribution networks, conventional voltage regulation methods struggle to address both overvoltage and undervoltage violations and to coordinate heterogeneous devices operating at different timescales. Existing region-based multi-agent control methods typically treat partitioned regions as homogeneous decision-making units and lack an explicit mechanism to characterize spatiotemporal variations in regional regulation capability and available flexibility. To address these limitations, this paper proposes a multi-timescale voltage optimization and control method for distribution networks based on regional priority evaluation and deep reinforcement learning. First, a coordinated voltage regulation architecture spanning long and short timescales is established. At the long timescale, a double deep Q-network (DDQN) is employed to centrally schedule discrete voltage regulation devices from a system-wide perspective. At the short timescale, a comprehensive regional voltage regulation evaluation index is introduced to quantify the regulation capability and potential of each region and thereby determine its regulation priority. A priority-aware state feature encoding (PASFE) module equipped with a learnable gating mechanism then adaptively fuses regional physical information with coordination information broadcast by the upper-level controller. The resulting priority-aware representations are incorporated into the regional actor and centralized critic networks, enabling the multi-agent algorithm to perform fine-grained regulation of individual continuously adjustable devices. Case studies on the IEEE 33-bus and 141-bus systems demonstrate that the proposed method effectively eliminates bus-voltage violations and outperforms the benchmark strategies in mitigating voltage deviations and reducing network losses, thereby confirming the effectiveness of regional-priority guidance in enhancing coordinated multi-timescale voltage regulation.
SN - 2045-2322
ER -