Design And Implementation Of State Management System For Satellite Application Capability Open Platform

With the increasing demands for satellite applications in various industries,the traditional approach of embedding applications in satellites cannot meet the business requirements anymore.Satellites need to possess application capabilities that are more scalable and extensible.The Satellite Application Capability Open Platform(SACOP)consists of the spacebased node and the ground-based node.The space-based node,i.e.,the satellite,utilizes container technology to enable flexible deployment of onboard applications and collaborates with the ground-based node deployed at the ground station to monitor the operational status of the satellite and the onboard application platform,as well as detect anomalies.This article designs and implements the System for managing the operational status in the Satellite Application Capability Open Platform(referred to as the "system").The system includes modules for data collection,processing,and transmission of operational status located in the space-based node,as well as modules for data persistence,anomaly detection,and status management located in the ground-based node.The system achieves monitoring of the infrastructure,container platform,and containerized applications’ operational status on the satellite.It addresses the high cost and latency issues of space-to-ground communication while enhancing the ground-based node’s capability to detect anomalies in the satellite’s operational status.This paper proposes a space-ground cooperative anomaly detection method(Space-Ground Cooperative Anomaly Detection Method,SGCAD)for communication-constrained environments.The method consists of three parts:operational metric data prediction and compression for the space-based node,and operational state anomaly detection for the groundbased node.In terms of time series data prediction for the space-based node,a Transformer-based approach is introduced.This method utilizes the attention mechanism of Transformer encoder layers as a variable in the model structure,transmitting it to the decoder.By extracting features at different scales through different encoding layers,the model learns deeper information and improves data prediction accuracy.Experimental results demonstrate that the proposed method outperforms baseline methods such as Long Short Term Memory(LSTM)and Transformer-Encoder.For time series data compression at the space-based node,an AutoEncoder-based method is proposed.This method models the AutoEncoder using two parallel LSTM units and a self-attention module,enhancing the AutoEncoder’s perception of time and reducing the distortion rate of important data,thereby improving data usability.Experimental results show that the proposed method achieves higher accuracy compared to baseline methods such as CNN-AutoEncoder and LSTM-AutoEncoder.In the ground-based node anomaly detection aspect,a continuous learning-based method is proposed.This method continuously updates the anomaly detection model by reusing similar pre-trained weights on new tasks,thereby improving anomaly detection accuracy across different tasks and avoiding catastrophic forgetting.Experimental results demonstrate that the proposed method achieves higher accuracy than baseline methods such as MAS.