Research On Satellite Power Parameters Abnormal Detection Method Based On SDAE

The Satellite Power System is one of the core components of the satellite.A satellite power failure can cause serious and irreparable losses.Anomalies in Satellite Power System is regarded as the "precursor" of the failure.If it is allowed to develop,it will lead to the complete failure of on-board mission,which requires continuous and accurate abnormal monitoring of satellite operation status.Since the ground control center’s grasp of satellite operation is completely from the sensor data of onboard condition monitoring system,with the continuous growth of satellite complexity and telemetry data,data-driven methods are used to analyzeand model the monitoring data,learn data characteristics,and realize abnormalities.The detection has become an important research direction in the aerospace field.Based on the characteristics of satellite power monitoring data,this paper proposes an anomaly detection method based on stacked denoising autoencoders(SDAE),which has important theoretical and application significance.This thesis firstly uses simulation methods to model the satellite power system,analyzes the abnormal characteristics of modules such as solar arrays,battery packs,and power controllers under normal and abnormal conditions.It is summarized that the simulated monitoring parameters have the characteristics of pseudo periodicity and mutual relevance.Secondly,based on the characteristics and variation rules of monitoring parameters,a SDAE data reconstruction model under a sliding window is proposed.The model is established by learning the characteristics of normal data,and then the parameters are are judged whether they are abnormal or not according to the reconstruction error.Finally,for the multi-dimensional monitoring parameters of the satellite power system,a multi parameter anomaly detection method based on data fusion is proposed.Calculate the Pearson correlation coefficient between parameters and group them according to the size of the correlation,use the principal component analysis method to reduce redundant information,and then build the SDAE data reconstruction network model from different channels.The calculated reconstruction error is judged by the Support Vector Data Description(SVDD),which detects abnormal data from normal data with the idea of single classification.This thesis uses public data sets to verify and evaluate the detection capabilities of the proposed algorithm for point anomalies and fragment anomalies.Compared with the traditional Autoregressive Integrated Moving Average(ARIMA)model and Long-short Term Memory(LSTM)model,the method in this paper reduces the false detection rate while achieving better system performance.Detecting the fragment anomalies of the satellite power system for single and multi-parameters can accurately find anomalies and achieve a higher recall rate.