Anomaly Detection Of Telemetry Data Based On Deep Learning

During the orbital operation of a spacecraft,a large number of telemetry data can be generated.These data can reflect the working status and operating conditions of various components during the operation of spacecraft in orbit,which is an important basis for spacecraft fault diagnosis and maintenance operation.Due to communication error or data sampling error and other reasons,telemetry data will have many uncertainties of its own,reliability and accuracy will be reduced.The low-reliability telemetry data will have a great impact on the following data applications,such as the analysis of spacecraft operation state and the analysis of the telemetry data based on big data.Therefore,effectively assessing the reliability of telemetry data is significant for spacecraft measurement and control.In this paper,the temperature and voltage parameters data of a certain spacecraft equipment are used as experimental data,and in-depth studies the data prediction based on deep learning.The credibility analysis of the telemetry data is based on the error of the prediction result,and finally realized abnormal detection of specific typed of telemetry data.The main work of this paper includes the following points: Analyzes the methods of time series anomaly detection and techniques and methods of sequence data prediction,understands the development status and research background;Analyze and study the deep learning related neural networks,selects the Long Short-Term Memory by comparison.The memory network model is used in the prediction of telemetry data;the ATTENTION mechanism is used as a combination of the long short-term memory networks as an enhance technology,and a P-LSTM prediction model is added with characteristic parameters,and experiments on the data set verify the effectiveness of the model;Based on the error between the predicted result and the actual data,the credibility analysis of the telemetry data was obtained,the corresponding credibility of the telemetry data was obtained,and the anomaly detection of the telemetry data was achieved.An anomaly detection verification model was designed to verify the anomaly detection model in this paper.Aiming at the problems of uncertainty and error code in the prediction model of telemetry time series data,as well as the problem of credibility analysis of time series data that has been proposed at present,experiments are carried out on real telemetry data sets.By comparing the experimental results are compared and demonstrated,it is proved that the anomaly detection model based on neural network telemetry time series data has practical significance.Finally,the advantages and disadvantages of the algorithm are summarized by analyzing the experimental results of the model.