基于GAN的GPS坐标时间序列降噪  

Noise Reduction of GPS Coordinate Time Series Based on GAN

作　　者：周新宇 陈刚 边家文 ZHOU Xinyu;CHEN Gang;BIAN Jiawen(School of Mathematics and Physics,China University of Geosciences,Wuhan 430074,China;College of Marine Science and Technology,China University of Geosciences,Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)数学与物理学院,武汉市430074 [2]中国地质大学(武汉)海洋学院,武汉市430074

出　　处：《大地测量与地球动力学》2025年第5期506-511,共6页Journal of Geodesy and Geodynamics

基　　金：国家自然科学基金(42274012)。

摘　　要：将生成对抗网络(generative adversarial networks, GAN)应用于GPS坐标时间序列的降噪处理中,通过生成器生成时间序列数据,并利用判别器对生成数据与干净的时间序列数据进行区分。在训练过程中,不断迭代优化生成器,直至判别器无法区分生成数据与真实干净数据,从而实现降噪效果。仿真实验结果表明,与WNNM、BiLSTM和SSA等传统方法相比,GAN方法在残差振幅、残差谱指数和速度不确定度等关键指标上与原始噪声的相关指标最接近,展现出更高效的降噪性能。真实数据的实验结果进一步验证了本文算法在实际GPS坐标时间序列降噪中的有效性。This paper applies generative adversarial networks(GAN)to the denoising of GPS coordinate time series.The generator produces time series data,which is then distinguished from clean time series data by the discriminator.During training,the generator is iteratively optimized until the discriminator can no longer differentiate between the generated data and the real clean data,thereby achieving the denoising effect.Simulation results show that,compared with traditional methods such as WNNM,BiLSTM,and SSA,the GAN method achieves the closest correlation with the original noise in key indicators such as residual amplitude,residual spectral index,and velocity uncertainty,demonstrating more efficient denoising performance.Experimental results using real data further validate the effectiveness of the proposed algorithm in denoising actual GPS coordinate time series.

关 键 词：GPS坐标时间序列 深度学习 GAN 信号降噪 

分 类 号：P228[天文地球—大地测量学与测量工程]