Abstract: The status monitoring, operational analysis, and anomaly detection of on-orbit satellites are critical to ensuring their safe and reliable operation, and play an important role in improving the efficiency of on-orbit management. Predicting the trend of satellite telemetry time series data is essential for anomaly detection and safe operation. With the increasing complexity of satellite systems and diversification of missions, the scale and dimensionality of telemetry data have rapidly expanded. Traditional prediction and detection methods based on empirical rules or simple statistical models struggle to meet the demands for efficient analysis and rapid response. Specifically, due to the lack of obvious periodic features in telemetry parameters, traditional methods face difficulties in effective modeling, resulting in insufficient prediction accuracy. This paper proposes a prediction model based on a linear-decoder combination. The model decomposes the time series to extract trend and seasonal components separately, and employs multi-granularity embedding strategies within the decoder to effectively capture both local and global temporal features. In the experimental section, the proposed method is applied to real telemetry data from an on-orbit satellite and compared against public time series forecasting datasets, demonstrating its feasibility and effectiveness.