IT Operations Automation · Predictive Maintenance
State Grid Fujian needed a province-level AI foundation that could standardize data collection, labeling, model development, deployment, and edge execution for power-grid inspection and operational use cases instead of leaving teams to work with isolated tooling.
Reported impact
Fujian standardized AI delivery across power-grid scenarios, improved defect detection and operational response, reduced manual reading and field risk, and created a reusable provincial AI foundation that can be copied across the State Grid system.
State Grid Fujian and Baidu AI Cloud built a province-wide 'two libraries and one platform' architecture to support AI in grid inspection and operations. The sample library centralizes labeled data from different professional domains, the model library manages self-built, third-party, and open-source models, and the platform provides training and runtime environments for machine learning, deep learning, and edge deployment. Baidu's AI middle platform connects requirements, samples, models, applications, and iteration so Fujian can run a full lifecycle from data extraction and annotation through training, evaluation, deployment, and edge distribution. The architecture supports notebook, visual, automated, and pipeline modeling while also enabling cloud-edge coordination through OpenVINO-based deployment tooling. It is used in scenarios such as UAV line inspection, substation operations, customer service, and power-grid monitoring, giving Fujian a reusable AI operating layer rather than one-off models for each business unit.
Operational and inspection data is drawn into the sample library, labeled and standardized, then used by the training platform to build and optimize models. Validated models are published into the model library and distributed through the AI middle platform and edge-deployment tooling to drones, substations, and other field or operational systems for live inference and alerting.
Average recognition across 16 defect scenarios reached 80%, frontline reading volume fell 90%, and substation operations improved 40-60x with energization time cut 80% (original: 16类场景缺陷平均识别率达80%，一线人员日均阅读量下降90%…效率比过去提升了40倍到60倍，启动送电时间缩短了80%).
Fujian built a 'two libraries and one platform' foundation covering sample resources, model resources, and training/runtime environments (original: 建设"两库一平台"智能基础设施…"两库"指模型库、样本库，"一平台"指包含训练环境和运行环境的人工智能平台).
The AI middle platform links requirements, samples, models, applications, and iteration into a full-process flow (original: 在样本库、模型库及运行环境的基础上实现需求、样本、模型、应用和迭代的全流程业务贯通).
The platform improved AI recognition accuracy by 30 percentage points and lifted recognition efficiency 5x in transmission-line inspection (original: 基于AI的智能识别算法将识别准确率提升了30个百分点，识别效率提升了5倍).
The province has collected 500,000 samples and developed 108 AI models (original: 福建电力已经归集样本50万条…研发各类人工智能应用模型108个).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Baidu AI Cloud Sample May 14, 2025 State Grid Hebei Electric Power Hebei Electric Power wanted a shared AI foundation that could stop duplicate AI investment, centralize data and model management and support many power-grid applications from monitoring to line inspection. AI Middle Platform Solution · AI Runtime Platform · AI Training Platform Baidu AI Cloud Sample May 15, 2025 Jinan No. 2 Machine Tool Group Co., Ltd. Jinan No. 2 Machine Tool needed to digitize production, quality inspection, maintenance and scheduling across a large CNC equipment business where fault monitoring, troubleshooting and planning still depended heavily on… AI-model development efficiency for safety scenarios improved about 30%. Predictive Maintenance Platform · Scheduling Optimization Engine · Smart Manufacturing Solution 
Microsoft Sample Jun 25, 2024 Furnas “An accelerator was needed, an architecture that gave us the possibility to accelerate deliveries.” Managing a large volume of data for a complex and critical system Predictive maintenance of transmission assets is… Having started in 2020, the deployment of the solution is expected to end in 2023. Azure · Azure AI · Azure Data Factory 