Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning

Abstract The increasing scale of unconventional oil and gas resource development has driven the demand for complex fracturing technologies. Due to their high value and energy consumption, the operational efficiency of fracturing units directly impacts the overall fracturing production efficiency. Cu...

Full description

Saved in:
Bibliographic Details
Main Authors: RuYi Wang, HuiShen Jiao, YingCheng Tian, Yi Zhao, SiQi Wang, Ke Zhang, Bo Huang, QinRui Sun, DanDan Zhu
Format: Article
Language:English
Published: SpringerOpen 2025-06-01
Series:Journal of Petroleum Exploration and Production Technology
Subjects:
Online Access:https://doi.org/10.1007/s13202-025-02014-7
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849333697152548864
author RuYi Wang
HuiShen Jiao
YingCheng Tian
Yi Zhao
SiQi Wang
Ke Zhang
Bo Huang
QinRui Sun
DanDan Zhu
author_facet RuYi Wang
HuiShen Jiao
YingCheng Tian
Yi Zhao
SiQi Wang
Ke Zhang
Bo Huang
QinRui Sun
DanDan Zhu
author_sort RuYi Wang
collection DOAJ
description Abstract The increasing scale of unconventional oil and gas resource development has driven the demand for complex fracturing technologies. Due to their high value and energy consumption, the operational efficiency of fracturing units directly impacts the overall fracturing production efficiency. Currently, the control of fracturing units mainly relies on manual experience, where experts allocate displacement for each fracturing truck based on total displacement requirements, which is then directly converted into pump speeds according to pre-set rules. However, this manual decision-making pattern often exhibits insufficient collaborative capability when confronted with complex construction scenarios. Furthermore, due to the lack of autonomous decision-making ability in concrete fracturing trucks, timely responses to unexpected failures are challenging. To address these issues, we proposes a multi-agent reinforcement learning based pump injection coordination control method to autonomously compute the optimal pump speeds for each fracturing truck in real time. The core of this method is an improved algorithm based on the TD3, which is enhanced by the incorporation of the CQL algorithm to improve the stability of the collaborative control strategy. A standardized pump injection control sequence dataset was constructed to verify the proposed method. And the experimental results illustrates that the average deviation between the calculated optimal pump speeds from the proposed method and the historical data is less than 8%, indicating that the proposed method can effectively mine the historical pump data and achieve expert-level control capability. Furthermore, the proposed method has been experimentally compared with both classical and cutting-edge models of machine learning and reinforcement learning, resulting in 37.5% to 48.6% reductions in pump speed deviation. It evidences the superiority of the proposed method within the realm of collaborative sequential decision making, showcasing its capability to achieve autonomous decision-making and coordinated control among multiple fracturing trucks. However, due to the limited data, the agents may have limited generalizability to different fracturing truck fleet configurations. Additionally, while the method shows promise in specific contexts, further online learning will be necessary for real-world deployment to adapt the agents to site-specific conditions.
format Article
id doaj-art-76559a07cb844943a4bb613ed76d184e
institution Kabale University
issn 2190-0558
2190-0566
language English
publishDate 2025-06-01
publisher SpringerOpen
record_format Article
series Journal of Petroleum Exploration and Production Technology
spelling doaj-art-76559a07cb844943a4bb613ed76d184e2025-08-20T03:45:47ZengSpringerOpenJournal of Petroleum Exploration and Production Technology2190-05582190-05662025-06-0115711510.1007/s13202-025-02014-7Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learningRuYi Wang0HuiShen Jiao1YingCheng Tian2Yi Zhao3SiQi Wang4Ke Zhang5Bo Huang6QinRui Sun7DanDan Zhu8Downhole Operation Research Institute, CNPC Engineering Technology R&D Company LimitedCollege of Artificial Intelligence, China University of Petroleum-BeijingDownhole Operation Research Institute, CNPC Engineering Technology R&D Company LimitedCollege of Artificial Intelligence, China University of Petroleum-BeijingDownhole Operation Research Institute, CNPC Engineering Technology R&D Company LimitedCollege of Artificial Intelligence, China University of Petroleum-BeijingDownhole Operation Research Institute, CNPC Engineering Technology R&D Company LimitedGWDC Engineering Technology Research InstituteCollege of Artificial Intelligence, China University of Petroleum-BeijingAbstract The increasing scale of unconventional oil and gas resource development has driven the demand for complex fracturing technologies. Due to their high value and energy consumption, the operational efficiency of fracturing units directly impacts the overall fracturing production efficiency. Currently, the control of fracturing units mainly relies on manual experience, where experts allocate displacement for each fracturing truck based on total displacement requirements, which is then directly converted into pump speeds according to pre-set rules. However, this manual decision-making pattern often exhibits insufficient collaborative capability when confronted with complex construction scenarios. Furthermore, due to the lack of autonomous decision-making ability in concrete fracturing trucks, timely responses to unexpected failures are challenging. To address these issues, we proposes a multi-agent reinforcement learning based pump injection coordination control method to autonomously compute the optimal pump speeds for each fracturing truck in real time. The core of this method is an improved algorithm based on the TD3, which is enhanced by the incorporation of the CQL algorithm to improve the stability of the collaborative control strategy. A standardized pump injection control sequence dataset was constructed to verify the proposed method. And the experimental results illustrates that the average deviation between the calculated optimal pump speeds from the proposed method and the historical data is less than 8%, indicating that the proposed method can effectively mine the historical pump data and achieve expert-level control capability. Furthermore, the proposed method has been experimentally compared with both classical and cutting-edge models of machine learning and reinforcement learning, resulting in 37.5% to 48.6% reductions in pump speed deviation. It evidences the superiority of the proposed method within the realm of collaborative sequential decision making, showcasing its capability to achieve autonomous decision-making and coordinated control among multiple fracturing trucks. However, due to the limited data, the agents may have limited generalizability to different fracturing truck fleet configurations. Additionally, while the method shows promise in specific contexts, further online learning will be necessary for real-world deployment to adapt the agents to site-specific conditions.https://doi.org/10.1007/s13202-025-02014-7Offline reinforcement learningHydraulic fracturingMulti-agent reinforcement learningMachine learning
spellingShingle RuYi Wang
HuiShen Jiao
YingCheng Tian
Yi Zhao
SiQi Wang
Ke Zhang
Bo Huang
QinRui Sun
DanDan Zhu
Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
Journal of Petroleum Exploration and Production Technology
Offline reinforcement learning
Hydraulic fracturing
Multi-agent reinforcement learning
Machine learning
title Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
title_full Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
title_fullStr Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
title_full_unstemmed Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
title_short Cooperative control method for multi-agent ground fracturing truck group based on offline reinforcement learning
title_sort cooperative control method for multi agent ground fracturing truck group based on offline reinforcement learning
topic Offline reinforcement learning
Hydraulic fracturing
Multi-agent reinforcement learning
Machine learning
url https://doi.org/10.1007/s13202-025-02014-7
work_keys_str_mv AT ruyiwang cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT huishenjiao cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT yingchengtian cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT yizhao cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT siqiwang cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT kezhang cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT bohuang cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT qinruisun cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning
AT dandanzhu cooperativecontrolmethodformultiagentgroundfracturingtruckgroupbasedonofflinereinforcementlearning