Article: Knowledge graph and reinforcement learning-based automatic decision-making for distribution network fault handling Journal: International Journal of Critical Infrastructures (IJCIS) 2026 Vol.22 No.14 pp.1 - 24 Abstract: To address the low decision accuracy, long decision time and limited fault recovery rate of traditional distribution network fault decision methods, this paper proposes an automatic decision-making method for distribution network fault handling based on knowledge graph and reinforcement learning. Firstly, the canopy clustering algorithm and UMAP manifold learning are adopted to collect and preprocess operational data of distribution networks. A top-down fault handling knowledge graph with pattern and data layers is further constructed to realise dynamic knowledge updating. Moreover, a knowledge graph-embedded deep Q-network decision framework is designed, which generates optimal fault handling sequences through offline dual-strategy learning. Experimental results verify that the proposed method achieves stable decision accuracy of 97.8-98.9% within 0.52-0.71 s, with a maximum fault recovery rate of 99.12%. Inderscience Publishers - linking academia, business and industry through research

Title: Knowledge graph and reinforcement learning-based automatic decision-making for distribution network fault handling

Authors: Wenshuo Lin; Jiawei Ai; Weixiang Zhang; Yiming Peng; Huili Tian; Chen Wang

Addresses: Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China ' Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China ' Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China ' Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China ' Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China ' Guangzhou Power Supply Bureau, Guangdong Power Grid Co., Ltd., Guangzhou, 510620, China

Abstract: To address the low decision accuracy, long decision time and limited fault recovery rate of traditional distribution network fault decision methods, this paper proposes an automatic decision-making method for distribution network fault handling based on knowledge graph and reinforcement learning. Firstly, the canopy clustering algorithm and UMAP manifold learning are adopted to collect and preprocess operational data of distribution networks. A top-down fault handling knowledge graph with pattern and data layers is further constructed to realise dynamic knowledge updating. Moreover, a knowledge graph-embedded deep Q-network decision framework is designed, which generates optimal fault handling sequences through offline dual-strategy learning. Experimental results verify that the proposed method achieves stable decision accuracy of 97.8-98.9% within 0.52-0.71 s, with a maximum fault recovery rate of 99.12%.

Keywords: distribution network fault handling; knowledge graph; reinforcement learning; canopy clustering; deep Q-network.

DOI: 10.1504/IJCIS.2026.157235

International Journal of Critical Infrastructures, 2026 Vol.22 No.14, pp.1 - 24

Received: 13 Jul 2026
Accepted: 28 Aug 2026
Published online: 23 Oct 2026 *

Title: Knowledge graph and reinforcement learning-based automatic decision-making for distribution network fault handling

Keep up-to-date