基于神经网络的准实时单站电离层TEC反演  

Near-real-time Ionospheric TEC Derivation from Single Station with Neural Network

作　　者：卢伟俊 马冠一[1,2] Lu Weijun;Ma Guanyi(National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100101, China;University of Chinese Academy of Sciences, Beijing 100049, China)

机构地区：[1]中国科学院国家天文台,北京100101 [2]中国科学院大学,北京100049

出　　处：《天文研究与技术》2022年第2期141-148,共8页Astronomical Research & Technology

基　　金：国家重点研发项目(2016YFB0501900);国家自然科学基金(11873064,12073049,12073049)资助.

摘　　要：使用全球导航卫星系统(Global Navigation Satellite System,GNSS)进行电离层总电子含量(Total Electron Content,TEC)反演时,单站相对于多站观测网是一种灵活简便的方法。基于人工神经网络(Artificial Neural Network,ANN)提出了一种准实时单站电离层总电子含量反演的方法。在这种方法中,上一个时段的硬件偏差作为初值并随着观测值调整,同时电离层总电子含量也准实时反演。为了对这种方法进行详细评估,通过位于中国的单站,4天磁静日的总电子含量分别采用本文方法与经典的最小二乘球谐函数法反演,其中硬件偏差和总电子含量的参考值通过附近的多站网得到。在另一个测试中,通过位于欧洲的单站,一次电离层暴事件及前后磁静日的总电子含量也分别通过上述方法反演。在磁静日,估计的硬件偏差整体上相对于最小二乘球谐函数法更接近参考值,反演的总电子密度更接近参考值。电离层暴时,两种方法反演的总电子含量也具有高度一致性,神经网络法估计的硬件偏差与磁静日的硬件偏差更接近。结果表明,提出的神经网络法相比最小二乘球谐函数法具有较高的精度。In derivation of ionospheric total electron content(TEC)by global navigation satellite system(GNSS),compared with a multi-station observation network,single-station derivation is a flexible and convenient method.Based on the artificial neural network(ANN),we propose a near-real-time method to derive ionospheric TEC with a single station.In this method,the instrumental biases of previous period are taken as initial values and are adjusted with the observation data.Meanwhile,the ionospheric TEC is derived in near-real-time.In order to have a detailed assessment on this method,through a single station in China,the TEC during four magnetically quiet days is derived by the proposed method and classic least square method(LSM)with spherical harmonics,respectively.The references of instrumental biases and TEC are obtained by the nearby multi-station network.In another test,through a single station in Europe,the TEC during an ionospheric storm and the quiet days before and after the event are also derived by the above methods.At the magnetically quiet days,the estimated instrumental biases are closer to the references than LSM with spherical harmonics on the whole,and the derived TEC is also closer to the references.During the ionospheric storm,the TEC derived by the two methods also have good consistency,and the estimated instrumental biases during the ionospheric storm are closer to those at magnetically quiet days.The results show that the proposed neural network method has higher accuracy than LSM with spherical harmonics.

关 键 词：电离层 人工神经网络 总电子密度 硬件偏差 电离层暴 

分 类 号：P228.4[天文地球—大地测量学与测量工程]