on the morning of december 27, 2023, the finals of the 5th southern power grid ai power dispatching application competition concluded at the southern power grid headquarters in guangzhou. the participating team from the department of electrical engineering and applied electronics (eea), tsinghua university, specializing in dispatching automation, secured the championship. additionally, the team swept the technical innovation award and best visualization award. the team members included doctoral students zeng hongtai and wang zhengcheng (on-site developer), and undergraduate students lu liangyuchen, deng kaihang, and ge yanshuo (remote participant), under the guidance of professor guo qinglai.
team members zeng hongtai and wang zhengcheng receiving awards on site
the southern power grid ai power dispatching application competition, organized by the southern power grid electric dispatch control center and the power system automation committee of the chinese society for electrical engineering, is currently one of the most influential and practical ai competitions in the power industry. the competition, themed “ai technology-based rapid clearance in the electricity spot market”, opened in guangzhou on november 14, 2023, with 30 teams participating from enterprises, universities, research institutions, and internet ecological companies across the country.
rapid clearance in the spot market requires solving large-scale unit combinations, essentially a mixed-integer linear programming problem. swift and accurate resolution of this issue has been a challenge in the industry. the participating team from eea adopted the “data-centric ai” technical route. initially, by tightening the feasible domain optimization, they reduced the average solving time for pure optimization by 55%, simultaneously lowering the solution tolerance by an order of magnitude, allowing them to generate higher-quality datasets in a shorter time. subsequently, they modeled unit start-stop predictions as a multi-task multi-class prediction problem, conducting supervised learning training on high-quality datasets to obtain an ai agent composed of deep neural networks. finally, the ai agent predicted the daily start-stop actions of units as integer variables for a fixed optimization problem, significantly reducing the solving time of the original optimization problem. moreover, the solution obtained approximated the original optimization result. this technical solution achieved a perfect score in both optimal values and solving time using real test data from the southern power grid, securing the team’s first-place position among the 30 participating teams.
finals on-site defense
eea’s dispatching automation research team has accumulated over thirty years of research expertise in the field of power grid control centers. it is one of the most influential research teams in china in the field of energy management and dispatch control. the team has received the national science and technology progress award (first class), national technology invention award (second class), and national science and technology progress award (second class) once each, and has been selected twice for the top ten scientific and technological advancements in chinese higher education. in recent years, the team has placed significant emphasis on the combined application of knowledge in power grid dispatching and artificial intelligence technology, achieving a series of important research results. currently, the team is undertaking several national key research and development projects, key projects of the national natural science foundation, and various practical engineering projects in related fields. the technical solution adopted by the team in this competition is one of the key research achievements of the southern power grid joint fund led by professor guo qinglai, titled “research on knowledge discovery of new type power system operation mode based on hybrid intelligence”.
