一种基于小数据量的UHF频段电离层闪烁事件人工智能预报新方法  

A new forecasting method of UHF-band ionospheric scintillations events by artificial intelligence based on a small dataset

作　　者：张红波[1] 王飞飞[1] 刘玉梅[1] ZHANG Hongbo;WANG Feifei;LIU Yumei(China Research Institute of Radiowave Propagation,Qingdao 266107,China)

机构地区：[1]中国电波传播研究所,青岛266107

出　　处：《电波科学学报》2023年第2期312-317,共6页Chinese Journal of Radio Science

基　　金：张明高院士工作室基金(A172011Y05)。

摘　　要：准确的电离层闪烁事件预警是空间天气预报的主要任务之一.针对中国低纬地区特高频(ultra high frequency,UHF)频段电离层闪烁事件预警信息需求,基于小数据量,充分利用经验知识和深度学习算法从电离层闪烁发生前的背景电离层参数中筛选有效的事件发生前兆因子,进而将电离层闪烁事件预报问题转换为观测数据的分类问题,最终基于深度信念网络形成了一种中国低纬地区UHF频段电离层闪烁事件预报新方法.利用该方法分析了多种观测数据组合与UHF频段电离层闪烁事件发生之间的相关性后,首次发现预报地区东侧跨赤道的电子总含量(total electron content,TEC)随纬度变化剖面的时序数据是电离层闪烁事件预报的重要前兆因子之一,对提升预报性能指标有显著帮助.Early warning of ionospheric scintillations is one of the main tasks of space weather forecasting.Faced with the forecasting information requirements of UHF-band ionospheric scintillation events over Chinese low-latitude region,the empirical knowledge and deep learning technique are used to identify the key precursors of ionospheric scintillations from the related ionospheric background parameters based on a small dataset.After that,the forecasting problem of post-sunset ionospheric scintillation events is converted to a classification problem easily solved by deep learning technique.Based on deep belief network of deep learning technique,a new forecasting method of UHF-band ionospheric scintillations events over Chinese low-latitude region has been established.After analyzing the correlations of different combinations of related observations and the occurrences of UHF-band scintillation events using this method,it is suggested that the latitudinal and daytime variations of the transequatorial total electron content(TEC)profile in the east of the forecasting area is one of the key precursors for forecasting UHF-band ionospheric scintillation events over Chinese low-latitude region,and is very helpful for the improvement of the forecasting performance.

关 键 词：电离层闪烁 事件预报 信息化作战 前兆筛选 经验知识 深度学习 

分 类 号：P352[天文地球—空间物理学]