Enhancing Power Transformer Fault Diagnosis Through Dynamic Knowledge Graph Reasoning
- 22 November 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-9
Abstract
Power transformers (PTs) are pivotal in electrical power systems as they facilitate transmission, distribution, and power supply by altering the ac voltage levels. Transformer failures can lead to power system interruptions and equipment damage, impacting the reliability and stability of power supply. Conventional fault diagnosis (FD) methods often struggle to effectively use fragmented knowledge and expert experience, resulting in reduced fault localization accuracy. To address these challenges, this article proposes a fault localization model for PTs using knowledge graphs (KGs). We present a novel knowledge extraction model to extract fact triples, facilitating the automatic construction of a PT fault KG (PTF-KG). In addition, existing KG-based FD methods often perform poorly in handling complex relationships and multihop reasoning. To overcome these limitations, we propose the knowledge graph reasoning model based on the dynamic interaction of dual agents (KGR-DIDA) model. The KGR-DIDA model comprises global and local agents, aiming to explore inference paths from source entities to target entities. The global agent uses a global policy to select the most relevant relationship, while the local agent, based on the chosen relationship, applies a local policy to navigate to the next entity, bringing us closer to the target entity. The action trajectory of the dual agent forms a reasoning chain that explains the fault query results. On the PTF-KG dataset, the KGR-DIDA model achieves an average FD accuracy of 97.5%. The experimental results validate that the KGR-DIDA model effectively addresses the challenges of multihop reasoning under complex relationships and significantly improves FD accuracy.Keywords
This publication has 24 references indexed in Scilit:
- FedSarah: A Novel Low-Latency Federated Learning Algorithm for Consumer-Centric Personalized Recommendation SystemsIEEE Transactions on Consumer Electronics, 2023
- AV4GAInsp: An Efficient Dual-Camera System for Identifying Defective Kernels of Cereal GrainsIEEE Robotics and Automation Letters, 2023
- Fault Diagnosis of Rolling Bearing Based on Knowledge Graph With Data Accumulation StrategyIEEE Sensors Journal, 2022
- Hierarchical Federated Learning for Power Transformer Fault DiagnosisIEEE Transactions on Instrumentation and Measurement, 2022
- GrainSpace: A Large-scale Dataset for Fine-grained and Domain-adaptive Recognition of Cereal Grains2022
- Transformer Fault Diagnosis Technology Based on Maximally Collapsing Metric Learning and Parameter Optimization Kernel Extreme Learning MachineIEEJ Transactions on Electrical and Electronic Engineering, 2022
- Representation iterative fusion based on heterogeneous graph neural network for joint entity and relation extractionKnowledge-Based Systems, 2021
- A Novel Self-Decision Fault Diagnosis Model Based on State-Oriented Correction for Power TransformerIEEE Transactions on Dielectrics and Electrical Insulation, 2020
- Relational retrieval using a combination of path-constrained random walksMachine Learning, 2010
- Simple statistical gradient-following algorithms for connectionist reinforcement learningMachine Learning, 1992