Integrated Design of 12 kV Intelligent Ring Main Unit: Advancing Intelligent Fault Diagnosis Through Multi-Model Fusion

Modern electrical distribution networks are expected to deliver increasingly reliable, resilient, and intelligent power services while accommodating distributed renewable energy resources and growing electricity demand. One critical component supporting these objectives is the 12 kV intelligent ring main unit (RMU), which plays an essential role in power distribution, fault protection, and operational monitoring. Conventional RMUs, however, often struggle with fragmented sensor information, delayed fault localization, and insufficient coordination among diagnostic subsystems. These limitations reduce fault diagnosis accuracy and prolong service restoration following electrical disturbances.

The reviewed study proposes an integrated intelligent RMU architecture that combines advanced hardware design with a novel multi-model artificial intelligence framework known as GAT-Transformer-DRSN. By integrating Graph Attention Networks (GAT), a spatiotemporal Transformer, and Deep Residual Shrinkage Networks (DRSN), the proposed system enables efficient multi-source information fusion, robust interference suppression, and rapid fault diagnosis. The research further validates the proposed framework using benchmark datasets and real operational data, demonstrating significant improvements in diagnostic accuracy and fault isolation performance while supporting the future development of intelligent distribution networks.

Bibliographic Information

Item	Information
Article Title	Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method
Authors	Rongrong Shan; Jieyun Qiu; Dong Han
Journal	Engineering Reports
Volume	8
Issue	1
Publication Year	2026
Article Number	e70548
DOI	https://doi.org/10.1002/eng2.70548
Publisher	John Wiley & Sons Ltd.
License	Creative Commons Attribution License (CC BY)
ISSN	2577-8196
Keywords	12 kV intelligent ring main unit; fault diagnosis; multi-source heterogeneous cross-modal information fusion; spatiotemporal transformer

1. Research Background

Distribution networks have become substantially more complex with the rapid deployment of distributed renewable energy resources, intelligent substations, and advanced automation technologies. Under these evolving operating conditions, traditional 12 kV ring main units are no longer sufficient to provide the level of monitoring, fault detection, and autonomous response required for modern power systems. Existing equipment generally relies on independently operating sensing devices and isolated monitoring subsystems, limiting the ability to integrate electrical, mechanical, and environmental information into comprehensive diagnostic decisions.

Current fault diagnosis methods also face significant technical challenges. Multi-source data generated by current transformers, voltage transformers, vibration sensors, temperature sensors, and environmental monitoring devices are often processed separately rather than collaboratively. This fragmented architecture reduces the capability to identify complex fault mechanisms, particularly those involving multiple interacting physical phenomena. Furthermore, conventional diagnostic algorithms frequently exhibit slow perception speed, inadequate cross-modal data fusion, and weak resistance to signal interference, leading to reduced diagnostic accuracy and delayed fault isolation.

The increasing penetration of distributed generation further complicates distribution network operation. Variable power flows, dynamic topology changes, and more diverse fault scenarios require intelligent equipment capable of understanding both spatial relationships among electrical components and temporal evolution of equipment conditions. Traditional threshold-based monitoring systems and single-model machine learning approaches are generally unable to capture these complex interactions effectively.

Previous studies have explored numerous approaches to improve distribution network fault diagnosis, including graph convolutional networks, Bayesian diagnostic frameworks, convolutional neural networks, dynamic Bayesian networks, graph learning, and knowledge graph-based methods. While these techniques have demonstrated improvements in specific scenarios, the reviewed article identifies several persistent limitations. Many existing methods cannot effectively integrate heterogeneous sensor information, perform dynamic cross-modal feature extraction, or simultaneously achieve high diagnostic accuracy, strong anti-interference capability, and rapid fault isolation.

Recognizing these limitations, the authors propose an integrated intelligent ring main unit architecture that combines hardware innovation with a multi-model artificial intelligence framework. Rather than relying on a single diagnostic model, the proposed solution integrates Graph Attention Networks (GAT), a spatiotemporal Transformer, and Deep Residual Shrinkage Networks (DRSN) into a unified fault diagnosis and isolation system. This integrated approach is intended to improve multi-source information fusion, accelerate fault localization, enhance robustness against signal interference, and establish a self-optimizing diagnostic workflow suitable for next-generation intelligent distribution networks.

2. Research Objective

To develop an integrated hardware and software architecture for a 12 kV intelligent ring main unit capable of supporting rapid fault diagnosis and isolation.
To design a multi-model artificial intelligence framework combining Graph Attention Networks (GAT), a spatiotemporal Transformer, and Deep Residual Shrinkage Networks (DRSN).
To improve multi-source heterogeneous data fusion by integrating electrical, mechanical, and environmental sensing information.
To enhance fault localization accuracy through topology-aware graph learning and spatiotemporal attention mechanisms.
To suppress signal interference using adaptive deep residual shrinkage techniques.
To establish a self-optimizing closed-loop fault diagnosis and isolation mechanism capable of supporting intelligent distribution network operation.
To experimentally evaluate the proposed framework using both benchmark datasets and real operational data collected from provincial power distribution networks.

3. Why This Research Matters

Supports next-generation smart distribution networks. The study contributes toward the modernization of electrical distribution infrastructure through intelligent monitoring and automated fault management.
Improves power supply reliability. Faster fault diagnosis and isolation reduce outage duration and improve service continuity for electricity consumers.
Integrates multiple sensing modalities. Electrical, mechanical, and environmental data are analyzed simultaneously rather than independently, enabling more comprehensive equipment assessment.
Introduces advanced artificial intelligence techniques. The combination of Graph Attention Networks, spatiotemporal Transformers, and Deep Residual Shrinkage Networks provides complementary capabilities that overcome many limitations of single-model approaches.
Enhances diagnostic robustness. Adaptive noise suppression improves reliability under challenging industrial environments where electromagnetic interference and signal distortion commonly occur.
Supports intelligent edge computing. The proposed architecture enables local processing and rapid decision making without relying entirely on centralized computing infrastructure.
Provides a practical engineering framework. The integration of hardware architecture, communication systems, and intelligent diagnostic algorithms demonstrates a comprehensive solution suitable for future industrial implementation.

4. Research Methodology

The study employed an engineering design and experimental validation approach that integrates intelligent hardware architecture with a multi-model artificial intelligence framework for fault diagnosis and isolation. Rather than improving only the diagnostic algorithm, the researchers simultaneously redesigned the sensing infrastructure, communication architecture, edge computing platform, and intelligent analytical models to establish a complete end-to-end diagnostic system for 12 kV intelligent ring main units.

The proposed framework consists of two major engineering components. The first is an integrated intelligent ring main unit architecture that combines multi-source sensing devices, high-speed execution mechanisms, and an edge computing gateway. The second is an intelligent fault diagnosis model that integrates Graph Attention Networks (GAT), a spatiotemporal Transformer, and Deep Residual Shrinkage Networks (DRSN) into a unified analytical framework capable of processing heterogeneous sensor information in real time.

Integrated Hardware Architecture

The intelligent ring main unit was designed around a hardware layer and a data transmission layer that operate collaboratively. The hardware layer integrates electrical parameter monitoring, mechanical condition sensing, environmental monitoring, and intelligent execution mechanisms. Electrical monitoring utilizes high-precision current transformers, voltage transformers, and zero-sequence transformers to detect overloads, short circuits, and grounding faults. Mechanical monitoring incorporates vibration and displacement sensors to identify switch mechanism abnormalities and equipment wear, while environmental monitoring includes temperature, humidity, and SF₆ gas concentration sensors to monitor insulation conditions and gas leakage.

The execution subsystem employs both electrically operated switching mechanisms and solid-state switches capable of millisecond-level fault isolation. To improve operational reliability under harsh industrial environments, the hardware architecture incorporates industrial-grade sensing components, active thermal management for power electronics, and wide-temperature embedded processors designed for continuous field operation.

Edge Computing and Data Transmission

A dedicated edge computing gateway performs protocol conversion, data preprocessing, feature extraction, local artificial intelligence inference, and secure communication. The gateway supports multiple industrial communication protocols while enabling real-time processing of heterogeneous sensor information. Data preprocessing includes wavelet-based denoising, feature compression, and standardized data representation before diagnostic analysis.

The architecture also incorporates multiple security mechanisms, including trusted device authentication, encrypted communication, access control, and secure execution policies to improve the reliability and cybersecurity of intelligent distribution network operation.

Multi-Model Artificial Intelligence Framework

The proposed GAT-Transformer-DRSN model combines three complementary deep learning techniques that address different aspects of intelligent fault diagnosis.

Graph Attention Networks (GAT) analyze the topological relationships among electrical components and dynamically identify fault propagation paths throughout the distribution network.
Spatiotemporal Transformer performs cross-modal information fusion by jointly modeling temporal evolution and spatial relationships among electrical, mechanical, and environmental sensing data.
Deep Residual Shrinkage Networks (DRSN) suppress environmental noise and electromagnetic interference while preserving weak fault features that are essential for accurate diagnosis.

The outputs generated by these three analytical components are subsequently fused to produce final fault diagnosis results and corresponding fault isolation decisions through a weighted decision-making mechanism.

Experimental Validation

The proposed framework was evaluated using two complementary datasets. The first dataset originated from the Electric Power Research Institute (EPRI), while the second consisted of operational fault records collected from a provincial distribution network in China. These datasets include multiple fault categories and heterogeneous monitoring information representative of practical distribution network operation.

Performance evaluation compared the proposed GAT-Transformer-DRSN framework with several existing intelligent diagnostic methods using commonly accepted classification metrics, including diagnostic accuracy, Macro-F1 score, false alarm rate, and fault recognition performance. Additional validation investigated computational efficiency, end-to-end processing delay, and hardware implementation feasibility on an edge computing platform.

5. Key Findings

Integrated Hardware and Artificial Intelligence Improve Fault Diagnosis

The proposed intelligent ring main unit architecture successfully combines integrated sensing, edge computing, and multi-model artificial intelligence into a unified diagnostic platform. The close collaboration between hardware and software enables rapid acquisition, processing, and interpretation of heterogeneous operational data, significantly improving overall diagnostic capability compared with conventional architectures.

Superior Diagnostic Accuracy

Experimental evaluation demonstrated that the proposed GAT-Transformer-DRSN model consistently achieved the highest diagnostic performance among all comparative methods. The framework obtained a Macro-F1 score of 0.9531, recognition accuracies of 96.28% and 96.45% on two independent datasets, and false alarm rates of only 0.75% and 0.72%, indicating highly reliable fault classification under diverse operating conditions.

Enhanced Multi-Source Information Fusion

The integration of Graph Attention Networks with the spatiotemporal Transformer enables effective fusion of electrical measurements, mechanical vibration signals, and environmental sensing information. Rather than analyzing each sensor independently, the framework dynamically captures relationships among heterogeneous data sources, improving the identification of complex fault mechanisms.

Accurate Fault Localization

Topology-aware graph learning significantly improves fault localization by dynamically modeling the physical relationships among distribution network components. The attention mechanism allows the diagnostic model to identify fault propagation paths more accurately than traditional graph-based approaches that rely on fixed adjacency relationships.

Robust Noise Suppression

The Deep Residual Shrinkage Network effectively suppresses electromagnetic interference, impulse noise, and environmental disturbances while preserving essential diagnostic information. This adaptive denoising capability substantially enhances the reliability of intelligent fault diagnosis in practical industrial environments where signal quality is often degraded.

Fast End-to-End Processing

Hardware implementation on an embedded Cortex-M7 edge computing platform demonstrated that the complete diagnostic workflow—including data acquisition, protocol conversion, artificial intelligence inference, and isolation command generation—can be executed within approximately 13.4 milliseconds. This rapid response satisfies the real-time requirements of intelligent distribution network protection.

Efficient Edge Computing Implementation

The experimental results show that advanced artificial intelligence models can operate efficiently on resource-constrained embedded hardware without requiring cloud-based processing. This capability enables intelligent ring main units to perform autonomous fault diagnosis locally while reducing communication delays and improving operational resilience.

Reliable Closed-Loop Fault Isolation

The proposed architecture establishes a complete closed-loop process in which sensing, diagnosis, decision making, execution, and verification operate collaboratively. Diagnostic outputs automatically trigger isolation commands that are subsequently verified through feedback mechanisms, improving both operational reliability and system safety.

6. Scientific Contribution

Introduces a comprehensive intelligent ring main unit architecture. The study integrates sensing technology, communication systems, edge computing, and artificial intelligence into a unified engineering framework rather than treating these components independently.
Develops a novel multi-model diagnostic framework. The proposed GAT-Transformer-DRSN architecture combines graph learning, spatiotemporal attention, and adaptive residual shrinkage to address complementary aspects of intelligent fault diagnosis.
Advances heterogeneous information fusion. The research demonstrates an effective strategy for integrating electrical, mechanical, and environmental sensing data within a single diagnostic model.
Enhances topology-aware fault analysis. Dynamic graph attention mechanisms improve fault localization by modeling evolving relationships among distribution network components.
Improves intelligent edge computing. The study validates that sophisticated deep learning models can operate efficiently on embedded hardware while satisfying strict real-time constraints.
Strengthens diagnostic robustness. Adaptive noise suppression enables reliable operation under challenging industrial environments characterized by electromagnetic interference and complex signal disturbances.
Provides practical engineering validation. The proposed framework is evaluated using both benchmark datasets and operational field data, demonstrating its applicability to real-world intelligent distribution networks.

7. Industrial Implications

Supports the modernization of electrical distribution networks. The integrated intelligent ring main unit architecture provides a practical solution for upgrading conventional distribution infrastructure toward smart grid operation, enabling faster fault detection, automated decision making, and improved operational reliability.
Improves power supply continuity. Rapid fault diagnosis and millisecond-level fault isolation reduce outage duration, minimize service interruptions, and enhance overall distribution network resilience, particularly in urban and industrial power systems.
Enables intelligent edge computing. By performing diagnostic inference directly on embedded edge devices, the proposed framework minimizes communication latency and reduces dependence on centralized cloud computing, making it suitable for real-time industrial applications.
Enhances predictive maintenance strategies. Continuous monitoring of electrical, mechanical, and environmental parameters enables early identification of equipment degradation before catastrophic failures occur, supporting condition-based maintenance rather than scheduled maintenance.
Reduces operational costs. More accurate fault localization decreases inspection time, shortens maintenance activities, minimizes unnecessary equipment replacement, and improves maintenance resource allocation.
Strengthens cybersecurity and operational reliability. Integrated authentication, encrypted communication, trusted data verification, and secure execution mechanisms improve the reliability and security of intelligent electrical infrastructure deployed in modern distribution networks.
Facilitates renewable energy integration. Intelligent distribution equipment capable of rapidly adapting to dynamic operating conditions supports the increasing penetration of distributed renewable energy resources while maintaining network stability.
Provides a scalable engineering platform. The modular architecture can potentially be extended to various intelligent switchgear, substations, feeder automation systems, and future smart distribution equipment.

8. Research Limitations

The study focuses specifically on 12 kV intelligent ring main units. The proposed architecture has not yet been validated for higher-voltage transmission equipment or other electrical network configurations.
Experimental validation relies on benchmark datasets and operational data collected from a provincial distribution network in China. Additional validation across different geographical regions and utility environments would further demonstrate the generalizability of the proposed framework.
The investigation primarily evaluates fault diagnosis and isolation performance. Long-term operational reliability, equipment aging, lifecycle maintenance, and economic performance are not comprehensively assessed.
The artificial intelligence framework integrates three advanced deep learning models, potentially increasing implementation complexity compared with conventional diagnostic systems.
Although the embedded implementation demonstrates real-time capability, the computational performance may vary depending on future hardware platforms, network sizes, and sensing configurations.
The proposed architecture emphasizes intelligent diagnosis under predefined fault categories. Its capability to identify entirely novel or previously unseen fault conditions requires additional investigation.
The study does not provide a detailed economic cost-benefit analysis for large-scale industrial deployment of the integrated intelligent ring main unit architecture.

9. Future Research Opportunities

Validate the proposed intelligent ring main unit architecture under larger and more diverse electrical distribution networks operating in different environmental and climatic conditions.
Investigate adaptive online learning techniques that enable diagnostic models to continuously improve as new operational data become available.
Expand the framework to support additional sensing modalities, including acoustic emission, infrared thermography, partial discharge imaging, and advanced power quality monitoring.
Develop lightweight artificial intelligence models that further reduce computational requirements while maintaining high diagnostic accuracy for embedded edge computing devices.
Investigate federated learning approaches that enable collaborative model improvement across multiple substations while preserving data privacy and cybersecurity.
Evaluate the proposed framework under cyber-physical attack scenarios to improve the resilience of intelligent power distribution systems against malicious interference.
Integrate digital twin technology with the proposed intelligent ring main unit architecture to enable real-time virtual monitoring, predictive simulation, and maintenance optimization.
Explore reinforcement learning strategies for autonomous fault recovery and adaptive distribution network reconfiguration following equipment failures.
Perform comprehensive economic assessments covering lifecycle cost, maintenance savings, energy efficiency improvements, and return on investment for utility operators.
Extend the proposed multi-model fusion framework to other intelligent electrical assets such as substations, transformers, switchgear, microgrids, and renewable energy integration systems.

10. Potential for Public Policy Citation

Although this study primarily addresses intelligent electrical engineering, its findings have considerable relevance for public policies supporting smart grid modernization, digital infrastructure, energy resilience, and sustainable electricity distribution. Governments and utility regulators worldwide increasingly emphasize the modernization of power distribution systems to accommodate renewable energy integration, improve service reliability, and strengthen infrastructure resilience. The proposed intelligent ring main unit architecture directly contributes to these strategic objectives by demonstrating how advanced sensing technologies, artificial intelligence, and edge computing can significantly improve operational performance.

The research also aligns with policy initiatives promoting digital transformation within critical infrastructure sectors. The integration of multi-source sensing, intelligent diagnostics, secure communication, and automated fault isolation illustrates practical implementation pathways for Industry 4.0 technologies in electrical utilities. Such technological capabilities support national strategies aimed at improving infrastructure reliability while reducing maintenance costs and operational risks.

In addition, the proposed framework contributes to broader energy transition policies by enabling more reliable integration of distributed renewable energy resources into existing distribution networks. As electricity systems become increasingly decentralized, intelligent monitoring and autonomous fault management will become essential components of future resilient energy infrastructure. Consequently, this research may serve as valuable technical evidence for policymakers, utility regulators, electrical standards organizations, and infrastructure planners developing future smart distribution network guidelines and investment strategies.

11. Who Should Read This Paper?

Power system engineers involved in distribution network planning, operation, and protection.
Researchers working in intelligent electrical systems, smart grids, artificial intelligence, and power system automation.
Electrical utility companies seeking advanced fault diagnosis and automated fault isolation technologies.
Industrial practitioners developing intelligent switchgear, ring main units, protection devices, and edge computing solutions.
Artificial intelligence researchers interested in graph neural networks, transformers, and multi-model deep learning applications.
Manufacturers of intelligent electrical equipment and digital substations.
Graduate students studying electrical engineering, intelligent systems, power electronics, or industrial artificial intelligence.
Government agencies and regulators responsible for smart grid modernization and critical infrastructure development.

12. Final Thoughts

This study presents a comprehensive engineering solution for intelligent fault diagnosis and isolation within modern electrical distribution networks. Rather than improving a single diagnostic algorithm, the authors successfully integrate advanced sensing technologies, secure communication infrastructure, edge computing, and multiple complementary deep learning models into a unified intelligent ring main unit architecture. The resulting GAT-Transformer-DRSN framework demonstrates excellent diagnostic accuracy, rapid response capability, and strong robustness against signal interference while remaining suitable for embedded real-time implementation.

One of the study's greatest strengths lies in its holistic engineering perspective. By simultaneously addressing hardware design, communication architecture, cybersecurity, intelligent data fusion, and embedded deployment, the research bridges the gap between theoretical artificial intelligence development and practical industrial implementation. The successful validation using both benchmark datasets and operational field data further strengthens the credibility of the proposed approach.

As electrical distribution systems continue evolving toward increasingly intelligent, decentralized, and renewable energy-intensive infrastructures, integrated diagnostic frameworks such as the one presented in this article will become increasingly valuable. The study therefore represents an important contribution to intelligent power distribution, smart grid engineering, and the broader application of artificial intelligence in critical infrastructure systems.

13. Suggested Citations

Teknomekanik (UNP) Style

Shan, R., Qiu, J., & Han, D. (2026). Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method. Engineering Reports, 8(1), e70548. https://doi.org/10.1002/eng2.70548

APA (7th Edition)

Shan, R., Qiu, J., & Han, D. (2026). Integrated design of 12 kV intelligent ring main unit with multi-model fusion and fault diagnosis method. Engineering Reports, 8(1), e70548. https://doi.org/10.1002/eng2.70548

IEEE Style

R. Shan, J. Qiu, and D. Han, "Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method," Engineering Reports, vol. 8, no. 1, Art. no. e70548, 2026, doi:10.1002/eng2.70548.

Harvard Style

Shan, R., Qiu, J. & Han, D., 2026. Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method. Engineering Reports, 8(1), e70548. Available at: https://doi.org/10.1002/eng2.70548.

Vancouver Style

Shan R, Qiu J, Han D. Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method. Engineering Reports. 2026;8(1):e70548. doi:10.1002/eng2.70548.

Chicago (Author–Date)

Shan, Rongrong, Jieyun Qiu, and Dong Han. 2026. "Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method." Engineering Reports 8 (1): e70548. https://doi.org/10.1002/eng2.70548.

MLA (9th Edition)

Shan, Rongrong, Jieyun Qiu, and Dong Han. "Integrated Design of 12 kV Intelligent Ring Main Unit With Multi-Model Fusion and Fault Diagnosis Method." Engineering Reports, vol. 8, no. 1, 2026, article e70548. Wiley, https://doi.org/10.1002/eng2.70548.

14. Editorial Note

Engineering Research Insights publishes independent scholarly reviews that summarize recently published engineering research for educational and scientific communication purposes. This review is intended to help researchers, engineers, graduate students, and practitioners quickly understand the scope, methodology, major findings, and practical significance of the reviewed study while encouraging readers to consult the complete original publication. Readers are strongly encouraged to read and cite the original peer-reviewed article whenever this work contributes to their research, teaching, engineering practice, literature review, or professional activities. Proper citation of the original publication acknowledges the authors' scientific contributions and supports responsible scholarly communication.

15. SEO Meta Description

Discover how a novel GAT-Transformer-DRSN framework enhances intelligent 12 kV ring main unit fault diagnosis through multi-source data fusion, edge computing, and advanced artificial intelligence for modern smart distribution networks.

16. SEO Keywords

12 kV intelligent ring main unit, intelligent ring main unit, ring main unit fault diagnosis, smart grid, electrical distribution network, Graph Attention Network, GAT, Transformer, spatiotemporal Transformer, Deep Residual Shrinkage Network, DRSN, multi-model fusion, multi-source data fusion, edge computing, artificial intelligence for power systems, intelligent fault diagnosis, electrical equipment monitoring, distribution automation, intelligent switchgear, Engineering Reports Wiley.