On August 10th 2023, the 23rd Expert Forum was conducted offline at the IUM meeting room while being livestreamed online at Zhihu and Koushare. Dr. Lv Baolei from Huayunshengda (Beijing) Meteorological Technology Co., Ltd. was invited to deliver a thematic report on Meteorological Environment Analysis and Research on Algorithm of Simulated Deep Learning Supervised by Physical and Chemical Models. Miao Shiguang, president of IUM, presided over the meeting with roughly 40,000 attendees, including directors and members of the association, researchers and graduate students of IUM, as well as personnel from relevant management, research, and business units across the country, which had an exposure of 5.76 million.
In Dr. Lv’s report, by establishing a mathematical description of physical and chemical processes for numerical models, deep learning can also realize the learning and restoration of data from the perspective of mathematical statistics. Dr. Lv has developed some deep learning methods with pattern data as constraints to achieve a multitude of functions such as generating meteorological environmental data analysis fields and simulating and predicting atmospheric environmental variables. In these deep learning models, the numerical solution process of partial physical and chemical equations can be explicitly expressed through deep learning modules, while the physical and chemical parameters and complex processes are implicitly learned through encoding and decoding modules. At the structural level, each module of the algorithm model is matched to the constraint equation set of each process. In this way, a great many functions such as reasonable parameter estimation analysis, internal process analysis, and sensitivity analysis can be achieved together with favorable simulation results.
All participants lively exchanged the report content through discussing the mathematical and physical methods of the study as well as the effect of these methods on the improvement of the numerical model. Evidently, Dr.Lv’s achievement provides guidance for the development of numerical models.
Video link: https://m.koushare.com/topic-hd/i/bjefms