On February 8, when seeing Chinese contestant Gu ailing win the gold medal of women's Freestyle Skiing Big Air with beautiful and difficult movements, Dr. Zhang Zhen of Chongqing Big Data Research Institute of Peking University couldn't help cheering. She pays more attention to the Beijing Winter Olympic Games than the ordinary audience, because the research results of her team are ensuring the weather forecast of the Beijing 2022 Winter Olympics.
Led by Zhang pingwen, academician of Chinese Academy of Sciences, vice president of Peking University and chief scientist of Chongqing Big Data Research Institute of Peking University, the research team of the Institute participated in the research of the key special project "research and application of objective prediction technology of fixed-point meteorological elements in the Winter Olympics" of the national key R & D plan "science and technology Winter Olympics", and developed an artificial intelligence MOML algorithm enabled weather prediction model to make the weather forecast more accurate.
Compared with the Summer Olympics, the Winter Olympics are more affected by the weather. Its meteorological support is the meteorological problem of medium and small-scale boundary layer under complex terrain conditions in winter. In order for the Winter Olympic athletes to play well on the field, they often require higher precision of prediction, even reaching the level of 100-meter and minutes, which has always been a difficulty in the international meteorological community.
"Our research is to revise the results of the weather forecast model through artificial intelligence algorithms to make it more accurate." Zhang Zhen is a doctor in the intelligent consultation and artificial intelligence weather forecast laboratory of Chongqing Big Data Research Institute of Peking University. She introduced that a large amount of meteorological data will be generated in meteorological business. At present, the numerical weather prediction model widely used in the world is to make numerical calculation through large-scale computers and express the physical process of weather evolution with physical equations, so as to predict the state of atmospheric motion and meteorology in a period of time. With the continuous improvement of global numerical weather prediction ability, it can basically solve the problem of large-scale prediction in most regions. However, for the demand of small-scale and refined prediction, there are errors in numerical calculation, and forecasters need to give prediction conclusions through consultation.
In the past, the consultation was highly dependent on forecasters, so it is necessary to synthesize the data of all parties and correct the deviation of model output data in combination with their own experience. The inherent advantages of artificial intelligence algorithm in information integration and processing can replace the process of forecasters integrating and analyzing information in consultation to a certain extent. Through data mining and learning, the forecasters' experience is internalized in the algorithm to realize intelligent and efficient prediction. The research team led by academician Zhang pingwen has developed the forecaster's artificial intelligence algorithm MOML, which realizes the intelligent correction, improves the prediction efficiency and further improves the prediction accuracy.
"Extensive research has been carried out within China and beyond on the correction method of mode output data deviation." Zhang Zhen introduced that the MOS method used previously is mainly revised for a single station. If people want to get an ideal correction result, they need to manually adjust the parameters, and the accuracy is limited. Through artificial intelligence algorithm, grid point prediction can be realized. At present, MOML algorithm has made a breakthrough in temperature, humidity, wind speed, wind direction and other weather elements. It can not only help forecasters and greatly reduce the workload of forecasters, but also improve the accuracy of prediction by more than 10% compared with conventional methods.
It is understood that the Beijing Winter Olympic Games has realized the "100-meter scale and minutes-level update" of short-term meteorological proximity forecast, which can quickly generate the objective analysis of weather elements such as temperature, humidity, wind and precipitation with a 100-meter resolution covering the Winter Olympic mountain stadium and updated every 10 minutes, as well as 0-12 hour forecast products.
Zhang Zhen said that in addition to serving the Winter Olympics, their team is also further studying the application of MOML algorithm in weather forecasting. In view of the more complex mountain environment in Chongqing, they are cooperating with Chongqing Meteorological Bureau to apply relevant research results in Chongqing.
