| 基于GAN序列的卫星数据接收系统故障定位 |
| |
| 作者姓名： | 王正晟 李亚林 张洪群 |
| |
| 作者单位： | 1. 中国科学院空天信息创新研究院, 北京 100094;2. 中国科学院大学电子电气与通信工程学院, 北京 100049 |
| |
| 基金项目： | 中国科学院战略先导科技专项A类(XDA15000000)资助 |
| |
| 摘 要： | 现有的卫星数据接收系统故障定位方法受困于专家知识的难以获取。提出一种基于生成对抗网络(GAN)序列的故障定位方法GANseq。利用故障传播特性，将接收系统分解为M个信号处理单元(SPU)，形成一个按照信号处理顺序排列的SPU序列；将故障定位问题转化为有序排列的M个异常检测子问题，其中第m个子问题为对前m个SPU的联合状态的异常检测；然后针对每个子问题采用基于GAN的状态参数检测模型，进而形成GAN序列；综合分析该序列在所有子问题上的检测结果以定位故障SPU。利用实际接收系统进行的实验证明所提出的GANseq不仅从数据驱动的角度完成了故障定位，而且在总体上降低了虚警率，并提高了定位精度。
|
| 关 键 词： | 卫星数据接收系统 故障定位 生成对抗网络 数据驱动 |
| 收稿时间： | 2020-10-30 |
| 修稿时间： | 2021-02-05 |
Satellite data receiving system fault location based on GAN sequence |
| |
| Authors: | WANG Zhengsheng LI Yalin ZHANG Hongqun |
| |
| Institution: | 1. Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;2. School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China |
| |
| Abstract: | Existing satellite data receiving system fault location methods suffer from the difficulty in obtaining expert knowledge. A generative adversarial network (GAN) sequence method called GANseq is proposed in this paper. According to the feature of fault propagation, the receiving system was first divided into M signal processing unit (SPU) to form a SPU sequence ranked by signal processing order (SPO). The fault location issue then was decomposed into ranked M anomaly detection sub-problems, where the mth sub-problem was the detection of the top m SPUs’ joint state. State parameters GANomaly-based detector was employed in each sub-problem, forming a GAN sequence. The detection results of this sequence on all sub-problems were analyzed to locate fault SPU. The experiment result from practical receiving system shows that GANseq can not only achieve fault location from data-driven perspective, but also reduce fault alarm rate and enhance accuracy level of fault location. |
| |
| Keywords: | satellite data receiving system fault location GAN data-driven |
|
| 点击此处可从《》浏览原始摘要信息 |
| 点击此处可从《》下载免费的PDF全文 |
