Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System

The increasing penetration of renewable energy sources (RESs) has led to the proliferation of small-scale distributed energy resources (DERs) in modern power systems. Effective coordination of these DERs in active distribution systems benefits both utilities and consumers. This paper introduces a no...

Full description

Saved in:
Bibliographic Details
Main Authors: Anurak Deanseekeaw, Watcharakorn Pinthurat, Boonruang Marungsri
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10979328/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850192076157747200
author Anurak Deanseekeaw
Watcharakorn Pinthurat
Boonruang Marungsri
author_facet Anurak Deanseekeaw
Watcharakorn Pinthurat
Boonruang Marungsri
author_sort Anurak Deanseekeaw
collection DOAJ
description The increasing penetration of renewable energy sources (RESs) has led to the proliferation of small-scale distributed energy resources (DERs) in modern power systems. Effective coordination of these DERs in active distribution systems benefits both utilities and consumers. This paper introduces a novel distributed active voltage and operation cost control (DAVOCC) framework designed to minimize node voltage deviations and operation costs. The proposed framework employs a multi-objective optimization approach, integrating three advanced algorithms: multi-agent proximal policy optimization (MAPPO), multi-agent asynchronous actor-critic (MAA2C), and multi-agent twin delayed deep deterministic policy gradient (MATD3). Battery energy storage systems (BESSs) and diesel generators (DGs) are used as heterogeneous agents, with their actions being constrained within predefined limits to ensure safe operation. The proposed framework is trained and tested on real-world data in a modified IEEE 33-node distribution system by featuring centralized training and decentralized execution (CTDE) framework. The obtained results demonstrate that three algorithms effectively maintain node voltage deviations within acceptable limits, with the MATD3-based algorithm achieving superior performance. Specifically, it delivers node voltage deviations close to nominal values, with an average deviation of 0.0042 p.u. and a standard deviation of 0.0065 p.u. Furthermore, the MATD3 algorithm reduces operational costs to 56,837.85 THB/day while generating the highest net profit of 725,943.71 THB/day from energy trading. These findings underscore the potential of the developed DAVOCC framework in optimizing the power management of BESSs and DGs, reducing the dependence on external grid energy and ensuring effective voltage regulation in active distribution systems.
format Article
id doaj-art-9531fbc836e34ec0b990870ddcc0c106
institution OA Journals
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-9531fbc836e34ec0b990870ddcc0c1062025-08-20T02:14:42ZengIEEEIEEE Access2169-35362025-01-0113756487566510.1109/ACCESS.2025.356512310979328Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution SystemAnurak Deanseekeaw0https://orcid.org/0009-0006-8940-5136Watcharakorn Pinthurat1https://orcid.org/0000-0002-0976-5873Boonruang Marungsri2https://orcid.org/0000-0002-7086-2197School of Electrical Engineering, Suranaree University of Technology, Nakhon Ratchasima, ThailandDepartment of Electrical Engineering, Rajamangala University of Technology Tawan-ok, Chanthaburi, ThailandSchool of Electrical Engineering, Suranaree University of Technology, Nakhon Ratchasima, ThailandThe increasing penetration of renewable energy sources (RESs) has led to the proliferation of small-scale distributed energy resources (DERs) in modern power systems. Effective coordination of these DERs in active distribution systems benefits both utilities and consumers. This paper introduces a novel distributed active voltage and operation cost control (DAVOCC) framework designed to minimize node voltage deviations and operation costs. The proposed framework employs a multi-objective optimization approach, integrating three advanced algorithms: multi-agent proximal policy optimization (MAPPO), multi-agent asynchronous actor-critic (MAA2C), and multi-agent twin delayed deep deterministic policy gradient (MATD3). Battery energy storage systems (BESSs) and diesel generators (DGs) are used as heterogeneous agents, with their actions being constrained within predefined limits to ensure safe operation. The proposed framework is trained and tested on real-world data in a modified IEEE 33-node distribution system by featuring centralized training and decentralized execution (CTDE) framework. The obtained results demonstrate that three algorithms effectively maintain node voltage deviations within acceptable limits, with the MATD3-based algorithm achieving superior performance. Specifically, it delivers node voltage deviations close to nominal values, with an average deviation of 0.0042 p.u. and a standard deviation of 0.0065 p.u. Furthermore, the MATD3 algorithm reduces operational costs to 56,837.85 THB/day while generating the highest net profit of 725,943.71 THB/day from energy trading. These findings underscore the potential of the developed DAVOCC framework in optimizing the power management of BESSs and DGs, reducing the dependence on external grid energy and ensuring effective voltage regulation in active distribution systems.https://ieeexplore.ieee.org/document/10979328/Multi-agent deep reinforcement learningmicrogridvoltage controloperation cost reductionbattery energy storage systemdiesel generator
spellingShingle Anurak Deanseekeaw
Watcharakorn Pinthurat
Boonruang Marungsri
Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
IEEE Access
Multi-agent deep reinforcement learning
microgrid
voltage control
operation cost reduction
battery energy storage system
diesel generator
title Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
title_full Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
title_fullStr Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
title_full_unstemmed Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
title_short Multi-Objective-Based Multi-Heterogeneous- Agent Deep Reinforcement Learning for Minimization of Voltage Deviation and Operation Cost in Active Distribution System
title_sort multi objective based multi heterogeneous agent deep reinforcement learning for minimization of voltage deviation and operation cost in active distribution system
topic Multi-agent deep reinforcement learning
microgrid
voltage control
operation cost reduction
battery energy storage system
diesel generator
url https://ieeexplore.ieee.org/document/10979328/
work_keys_str_mv AT anurakdeanseekeaw multiobjectivebasedmultiheterogeneousagentdeepreinforcementlearningforminimizationofvoltagedeviationandoperationcostinactivedistributionsystem
AT watcharakornpinthurat multiobjectivebasedmultiheterogeneousagentdeepreinforcementlearningforminimizationofvoltagedeviationandoperationcostinactivedistributionsystem
AT boonruangmarungsri multiobjectivebasedmultiheterogeneousagentdeepreinforcementlearningforminimizationofvoltagedeviationandoperationcostinactivedistributionsystem