23-03 2026
AI Fault Location Assistant for Distribution Networks Launched, Cutting Location Time by 58%
Kunming Power Supply Bureau of Yunnan Power Grid Company has recently developed an AI fault location assistant for distribution networks. Leveraging local computing power and fault recording technology, the assistant integrates multi-source data through AI algorithms to achieve intelligent fault identification, precise location, and automated information push across the entire process. The fault type identification accuracy rate reaches 87.6%, while the average time required for fault location has been reduced from 4.32 hours under traditional maintenance to 1.81 hours, a decrease of 58%. This provides an efficient and reliable fault handling solution for the "urban + rural + mountainous" hybrid distribution network in the Yunnan region.
The distribution network in the Kunming area covers a wide range with a complex structure, encompassing densely configured urban lines, decentralized rural lines, and lines in complex mountainous terrain. Data shows that under the traditional model, the average time for fault location in the Kunming area was approximately 4.32 hours, with fault location on mountainous lines taking more than twice as long as in urban areas. The fault handling model had long faced the dual challenges of efficiency and accuracy.
"In the past, constrained by technological means, most fault recording data from distribution networks was not effectively utilized. With the improvement of distribution automation and protection information systems, fault recording data now has remote retrieval capabilities and can support decision-making in handling faults," said Hu Zejiang, Deputy Manager of the Power Dispatch Control Center at Kunming Power Supply Bureau.
How to activate this "dormant" fault recording data became the core issue in breaking through the bottlenecks of distribution network operation and maintenance.
To address the pain points of the traditional model, the Power Dispatch Control Center of Kunming Power Supply Bureau, leveraging its Yang Pengjie Relay Protection Innovation Studio and Zhou Yanping Dispatch AI Innovation Studio, formed a dedicated team in collaboration with Yunnan Electric Power Research Institute. Relying on local computing resources, they creatively constructed a technological system centered on "multi-source data fusion + AI algorithm modeling + intelligent information push" to develop the AI fault location assistant for distribution networks. This system realizes three core functions: intelligent identification of fault types, automatic calculation of fault distances, and automatic push of fault information.
The team deeply applied AI algorithms to construct an automatic fault type discrimination model. By calling upon fault recording data from distribution automation terminals and 10 kV indoor station switches, the model automatically classifies, labels, and deeply analyzes key information such as waveform characteristics, amplitude changes, and frequency offsets, accurately identifying fault types and phases, including short circuits and ground faults.
"The model's training utilized over 2,000 fault cases that occurred in recent years, employing a multimodal deep learning framework to analyze and label characteristic waveforms, ensuring the algorithm adapts to the differentiated fault characteristics of various lines," introduced Wang Honglin, a Senior Researcher in Distribution Technology at Yunnan Electric Power Research Institute. The team, by integrating topology analysis algorithms, achieved, for the first time, the deep fusion of fault recording data with distribution network geographic and real-time operational data, establishing a comprehensive fault feature library covering "urban-rural-mountainous" scenarios.
Addressing the differentiated fault characteristics of the hybrid distribution network in the Yunnan region, the team innovatively introduced a dynamic threshold adjustment mechanism. Combined with topology structure and real-time data, the system automatically utilizes core information such as line parameters, equipment PTs (potential transformers), and CTs (current transformers) to calculate fault distances. After multi-dimensional cross-validation, the fault location accuracy has been improved by 40% compared to traditional methods. "Previously, manual calculation of fault distances had significant errors and was time-consuming. Now, the AI assistant can pinpoint the location within seconds, greatly reducing the burden on grassroots staff," said Chen Tao, Assistant Manager of the Planning and Production Department at Kunming Dongchuan Power Supply Bureau of Yunnan Power Grid Company.
Technological innovation has not only improved fault handling efficiency but also facilitated a role transformation for operation and maintenance personnel. "Now dispatchers are no longer overwhelmed by massive amounts of information but can focus on core decision-making for fault handling. Maintenance personnel can also devote more energy to equipment status analysis and hidden danger investigation, achieving reduced burden and increased efficiency for both dispatchers and maintenance staff simultaneously," Hu Zejiang stated. He noted that the new model realizes an optimal combination of "AI pre-judgment + human decision-making," making distribution network operation and maintenance smarter and more efficient.
"As new types of power equipment, such as distributed photovoltaics and energy storage, are increasingly connected to the distribution network, fault characteristics will become more complex. The self-learning capability of the AI fault location assistant will provide crucial support for the safe and stable operation of the new type power system," Hu Zejiang added.
