Zhou Pan: Empowering power production with big models

On May 10, the Central Enterprise AI+Large Model Application Forum was held in Beijing, hosted by the China Enterprise Reform and Development Research Association, the Digital Economy and Data Assets Working Committee of the China Enterprise Research Institute, and the Enterprise Observer. At the forum, Zhou Pan, a solution expert at the Southern Power Grid, shared his views on the practice of large-watt large-model power production applications of the Southern Power Grid.

Zhou Pan said that based on the business architecture and capability list of the power production domain, China Southern Power Grid , in conjunction with business experts , adopted a divergent-first-then- convergent strategy to sort out the needs of artificial intelligence for various businesses . In terms of data fusion, status assessment, image recognition , content generation, etc., a total of 21 key AI+ construction scenarios have been sorted out, focusing on intelligent inspections, intelligent maintenance, equipment status evaluation, etc., and 13 AI+ solutions have been planned.

At the level of overall technical solutions, the State Grid Corporation of China takes "Big Watt" CV, NLP, multimodality, and scientific computing large models as the core, integrates the core data of the digital power grid, builds a high-quality data sample set in the production domain, integrates the professional knowledge base in the production domain , and unifies the management of power production algorithm models. It utilizes the task orchestration and collaborative calling capabilities of the intelligent platform and opens up the interface between edge reasoning and the production operation support system , completes pilot research and development applications for scenarios such as production domain command decision- making, equipment operation and maintenance optimization, meteorological warning and forecasting for disaster prevention and mitigation, disaster assessment, and smart ticketing, to support intelligent perception , operation , and decision-making in the production domain .

The intelligent inspection solution of the Southern Power Grid is driven by the dual wheels of "business + AI technology". Through "reuse + integration + innovation", it quickly forms service capabilities, forms a large model of L2 transmission and distribution inspection scene level, and completes a complete set of intelligent inspection solutions with intelligent recognition algorithms as the core. In terms of power transmission and distribution inspection, cameras and drones equipped with intelligent algorithms are mainly used to carry out auxiliary inspections, and combined with image recognition technology, mountain fires, component defects, line foreign objects, channel hazards, etc. are intelligently identified to achieve unmanned inspections, fully explore the value of inspection data, improve the utilization rate of inspection data, and enhance the intelligent level of transmission and distribution business applications. In addition, the Southern Power Grid has established a new model for the application of intelligent inspection of transmission lines, using intelligent bodies for intelligent analysis, judgment, and decision-making, and carrying out automatic inspections, automatic identification, and automatic analysis reports throughout the on-site inspection process, realizing defect status prediction and early warning multi-task collaborative processing, achieving a power inspection machine replacement rate of more than 75%, and realizing the automatic arrangement of inspection results. Operation plan, follow-up on the progress of defect elimination, full business unmanned.

On the large model of power grid maintenance, the Southern Power Grid has trained the L2 transmission, transformation and distribution network maintenance scenario large model based on power industry knowledge, Southern Power Grid power safety work procedures, business management requirements, single/ticket templates, typical tickets/typical reports and other data, and strengthened the large model's capabilities in the field of power grid maintenance business, such as association analysis, feature analysis, logical reasoning, and knowledge question and answer. At the same time, it integrates multi-source data inputs such as equipment ledgers, technical parameters, power grid topology, operating status, and timing measurement, and combines existing software capabilities such as topology analysis and power flow calculation to achieve auxiliary maintenance decisions such as maintenance strategy generation and maintenance plan scheduling, and realizes the automatic generation of maintenance plans and operation documents such as maintenance orders, work tickets, operation tickets, emergency plans, and fault analysis reports.

In terms of intelligent solutions for disaster prevention and reduction, China Southern Power Grid implements the 15-character policy of "pre-disaster prevention, pre-disaster prevention, defense during disasters, post-disaster rescue, and post-event evaluation". Based on the "Big Watt" framework, it formulates an application solution of "large and small models + cloud-edge collaboration" for disasters such as wildfires, typhoons, geology, and rainstorms. It builds seven types of disaster prevention and emergency scenarios such as typhoon disaster disposal and icing disaster disposal from multiple dimensions of "before, during, and after the disaster". It combines video and IoT sensing data to conduct a comprehensive risk assessment of the operating status of the transmission line, transforming the nodes of manual intervention in emergency disposal into human-machine collaboration and machine-machine collaboration to improve emergency efficiency. Through the training and research of AI meteorological models in the production domain, China Southern Power Grid has now achieved accurate forecasts of dangerous weather such as typhoons, heavy rainfall, high temperatures, and cold waves. Relying on the multi-time scale intelligent prediction model of disasters such as icing, waterlogging, geological disasters, and wildfire spread at the power facility level, it has improved the level of production command and foresight, and transformed the prevention and control of operational risks from passive response to active defense.

Zhou Pan introduced that the power transmission artificial intelligence large model (Big Watt CV) has been applied on a large scale in various cities in Guangxi. The model can identify 35 types of defect hidden dangers such as fire risks. Since 2025, the cumulative call volume has reached 77.04 million times, analyzed 11.585 million pictures, effectively identified 199,100 power transmission defects, and identified 11,400 fire information in total. 8,244 monitoring fire point alarm information was verified and handled, and 783 fire points that may endanger the safe operation of the line were controlled. The defect identification accuracy rate reached 85%, which effectively promoted the large-scale application of the "drone autonomous inspection + defect intelligent identification" intelligent operation mode, and provided strong technical support for mountain fire prevention and control and power supply equipment operation and maintenance.