基于条件扩散模型的卫星遥测数据缺失值插补方法

Satellite telemetry time-series data is critically valuable for applications like remote sensing monitoring and navigation positioning, and is also effective for monitoring satellite health status. However, these data often suffer from missing values caused by complex factors such as sensor malfunctions and data transmission errors, which severely impair data integrity and usability, potentially leading to erroneous decisions. To address this, a satellite time-series completion method based on multivariate conditional diffusion model is proposed to improve the accuracy of missing value imputation in satellite telemetry data. The method first incorporates a conditional diffusion method that uses observed satellite data as conditional input to generate data by modeling the posterior distribution of the missing values. Preliminary linear imputation is applied to the incomplete samples during generation to enhance the stability of the model. Furthermore, a residual module integrating a temporal attention layer and a gated activation unit serves as the core prediction network, effectively capturing temporal dependencies in the multivariate telemetry data for precise reconstruction of missing values. Finally, Extensive experiments were conducted on a momentum wheel telemetry dataset from a communications satellite and on public time-series datasets. The experimental results show that the proposed method delivers robust performance and generalization ability across varying missing rates, outperforming existing methods in both accuracy and stability.