Only 7 days to develop a set of AI applications, the power system bus load forecasting accuracy increased by 4 percentage points – this is the Guangdong power grid power control center staff in the AI application competition results.
“Based on the cloud platform built by China Southern Power Grid and Alibaba cloud, ordinary dispatchers can learn and use it now. Through drag and drop and simple code, they can achieve AI load forecasting, and the accuracy rate is even higher than that of computer department students, which is unimaginable in the past.” Said Liang Shouyu, a first-class technical expert and professor level senior engineer of China Southern Power Grid.
On January 28, the second power dispatching AI application competition was successfully closed in the research base of China Southern Power Grid. With the help of AI algorithm, 45 teams from universities, scientific research institutions and enterprises challenge the bus load forecasting problem in the industry. Finally, the prediction accuracy of 25 teams is higher than that of Zhanjiang mode. Yike energy won the championship with a total score of 73.9 points (85% accuracy), and its comprehensive efficiency was 10% higher than that of manual operation.
The competition is jointly held by the power dispatching and control center of China Southern Power Grid and the power system Automation Committee of China Electrical Engineering Society. It is the most influential and applied AI competition in the power industry. Previously, the whole network load forecasting scheme won in the first AI competition has been applied to power grid business, and the accuracy rate has been increased to 97.63%. It is expected that it will be fully used this year.
Compared with the whole network load forecasting, it is more difficult to forecast the bus load in Zhanjiang. China Southern Power Grid has 17 220 kV power stations and 33 220 kV buses in Zhanjiang, Guangdong Province. Every 15 minute interval means that 3168 load forecasting values need to be predicted every day. On the other hand, the bus power supply range has a direct impact on the accuracy, and the change of power consumption in a steel plant may cause a sudden change in the load curve.
In order for AI to make accurate prediction, first of all, it needs machine learning historical data. This is a massive database, covering the load curve, grid structure, historical maintenance ticket data, meteorological data, etc It can take years for one person to go through it one by one.
The dispatching cloud platform built by China Southern Power Grid Based on Alibaba cloud technology has come into use.
Power dispatching is the core system in the power grid. It is necessary to arrange the plans of all power plants and users to ensure the real-time balance of power generation and consumption. The power dispatching of China Southern Power Grid covers five provinces, and the system is complex. In 2018, China Southern Power Grid took the lead in building the “brain center” of power grid system dispatching cloud platform in the industry, and officially launched in August 2019. Based on the powerful computing power of the dispatching cloud, the computing time of the day ahead security check function of the South Power Grid Dispatching Center has been shortened from the past hour to 10 minutes.
In addition to stability and efficiency, scheduling cloud is also an open technology development platform. “The scheduling cloud is equivalent to the operating system of mobile phones and the ecology of development and operation. We can quickly develop various applications on the cloud, just like the app of mobile phones.” Liang Shouyu made an analogy.
Only three months after the completion of the dispatching cloud platform, China Southern Power Grid quickly launched the “electricity spot trading” business. Through the automatic allocation of supply and demand relations and market rules, the electricity prices of different regions and time periods are dynamically adjusted every 15 minutes, so as to minimize the electricity cost of the whole society. According to the data, the daily electricity charge of a large power user has dropped from 190000 to 78000.
To some extent, AI competition is the ultimate challenge to the agile development ability of cloud platform. In the past, it took more than a year for such load forecasting projects to build platforms, process data, and launch applications. But the AI contest only gives developers seven days, and they have to speed things up.
The scheduling cloud platform first provides a cloud desktop terminal for contestants to access massive historical data at any time. The machine learning Pai platform built on the scheduling cloud also provides a large number of “drag and drop” algorithm training tools to support modular “low code” development. At the same time, through products such as dataworks, quickbi and maxcompute, the data processing is greatly accelerated to help developers screen out the appropriate algorithm model as soon as possible.
“The original independent research and development is not difficult,” said Zheng Wenjie, a contestant of Guangdong power grid power dispatching center. “In fact, in the actual work of power grid dispatching, there are many business scenarios that can be quickly iterated and developed through this platform. The competition is over, but we will continue to push forward.”
The dispatching cloud platform is still expanding, in order to help more innovation in the power industry, and truly make electricity safer and cheaper.
Alibaba cloud and other technical support expansion projects have been started. It is expected that after the expansion is completed, more than 50% of the business systems of China Southern Power Grid will run on the dispatching cloud. Based on the powerful and stable basic cloud platform, the new server of the scheduling cloud can be plug and play, and the resources used by the application can be expanded in seconds. In the past, it took several months.
Based on the dispatching cloud platform, a new weather forecasting supercomputing platform will also be launched before the first typhoon of this year, so as to reduce the impact of bad weather on power consumption. This is also the first time in the power industry.
(editor in chief: Shen Jiaping, LV Qian)