Recently, Southern Power Grid Company's first independently controllable AI mega-model for power distribution networks (hereinafter referred to as the distribution AI mega-model) was launched and applied in a physical form at Guangzhou Power Supply Bureau of Guangdong Electric Power Company. This marks the first engineering application focused on intelligent inspection scenarios in the power distribution field since the Southern Power Grid Company introduced the "Big Watt" (AI model).
The distribution AI mega-model is applied to the production and operation support system (edge side) of the power distribution network. For the first time, the power distribution AI model is deeply integrated into the inspection business process and operational steps, further deepening the application of artificial intelligence technology in equipment status evaluation, defect and hidden danger diagnosis, and prevention of external damage. The traditional model of personnel reporting inspection results layer by layer will gradually transition to a new intelligent inspection model characterized by scalability, precision, and reduced manpower, empowering frontline production staff to work more efficiently. Taking the inspection of a single 10 kV feeder line as an example, the average workload under the traditional model is 8 person-hours of on-site inspection. Under the new model, it is expected to reduce to 0.2 person-hours of back-office manual review, decreasing the inspection duration by 97.5% and significantly improving the efficiency of power distribution network inspection.
It is understood that the distribution AI mega-model is based on the 3 billion parameter foundation of the "Big Watt" L1 level of the Southern Power Grid Company. During the development process, the Guangzhou Power Supply Bureau focused deeply on sample supply, computing power construction, and model optimization.
As of April 30, district bureaus such as Yuexiu, Haizhu, Huangpu, Nansha, and Panyu in Guangzhou have started trial use of the distribution AI mega-model. They have carried out 6 inspection tasks, achieving an overall abnormality recognition rate of over 80% for issues like water seepage and small animal intrusions. Additionally, the distribution AI mega-model can efficiently identify typical defects and hidden dangers in the power distribution network, such as bird nests, hanging objects, damaged porcelain insulators, and missing cable ties, providing higher quality, more reliable, and more convenient AI technical support for the "machine inspection replacing human inspection" in the power distribution network.
