The current satellite management system mainly relies on manual work. If small faults cannot be found in time, it may cause systematic fault problems and then affect the accuracy of satellite data and the service quality of meteorological satellite. If the operation trend of satellite will be predicted, the fault can be avoided. However, the satellite system is complex, and the telemetry signal is unstable, nonlinear, and time-related. It is difficult to predict through a certain model. Based on these, this paper proposes a bidirectional long short-term memory (BiLSTM) deep leaning model to predict the operation trend of meteorological satellite. In the method, the layer number of the model is designed to be two, and the prediction results, which are forecasted by LSTM network as the future trend data and historical data, are both taken as the input of BiLSTM model. The dataset for the research is generated and transmitted from Advanced Geostationary Radiation Imager (AGRI), which is the load of FY4A meteorological satellite. In order to demonstrate the superiority of the BiLSTM prediction model, it is compared with LSTM based on the same dataset in the experiment. The result shows that the BiLSTM method reports a state-of-the-art performance on satellite telemetry data.

Parallel Application Scheduling Model Based on Petri Net with Changeable Structure | InTechOpen, Published on: 2010-09-27. Authors: XianGang Zhao, CaiYing Wei, ManYun Lin, et

Satellite housekeeping systems combine heterogeneous sensor channels that require continuous fault monitoring. Existing methods rely mainly on hand-crafted statistics or generic raw-signal models, which can obscure localized morphology and do not explicitly adapt shared features to sensor type. They also rarely assess accuracy together with resource cost and temporal evidence, limiting deployment relevance. To address these limitations, we present OpsSatNet, a lightweight sensor-aware attention network for segment-level anomaly detection from raw, resampled telemetry. Specifically, a dilated residual convolutional backbone captures multi-scale patterns, the feature-wise linear modulation adapts shared features by sensor type, and additive attention pooling aggregates temporal features with learned weights. On OPSSAT-AD, the base model achieves an F1-score of $0.960\pm 0.009$ over five runs, compared with $0.946$ for the strongest published baseline and $0.949\pm 0.004$ for the strongest retrained raw-signal baseline. It contains 258K parameters, while a 16K parameter variant retains an F1-score of $0.952\pm 0.010$ at 8 MFLOPs and 1.157 ms single-thread CPU latency per segment. Ablations show positive mean F1 differences for sensor conditioning and attention pooling, although neither individual effect is statistically significant. These results support lightweight, sensor-aware monitoring, while cross-mission and flight-hardware validation remain necessary.