A Novel Method for Satellite Monitoring With One-Dimension Feature Based on Autoencoder Model

Abstract

In order to monitor all telemetry data, thresholds are adopted to judge the status of satellite. This method is terrible when some abnormal happened, if the data was not more than pre-set threshold. when the data exceeding the threshold after a period of time, there were a big fault for satellite. This fault would make a huge economic loss especially for the communicate satellite. These are two classes telemetry of satellite about this scenario, one class is continuously changing digital telemetry, the other class is temperature. A method was proposed for solving these problems. An autoencoder model was applied to monitor the telemetry data according to the devices or equipment board. Each device or equipment board has own model, and telemetry data is inputted to the model for compressing a single parameter as one-dimension feature. The operators just only monitor the one-dimension feature, that is simple and fast. If an abnormal appear, the parameter of device or equipment board would be changed to warn the operators, who would check the actual telemetry data of device or equipment board, and the abnormal would be checked out immediately and earlier than the traditional method. For detecting the two kinds of typical abnormal which could not detect by traditional method, two models were built and data was prepared. The results show that auto-decoder model can detect the abnormal accurately and be useful for the operator. A software was built, and some models were trained for a satellite.