Research On Fault Simulation And Diagnosis Of GEO Satellite Power System

The satellite power system is the only energy source of the satellite,and it is also one of the sub-systems that are very prone to failure.Improving its repair and maintenance capabilities through effective fault diagnosis technology has always been the focus of research in the field of satellite power.On the one hand,due to economics,repeatability and other reasons,most of the fault diagnosis research on satellite power supply adopts simulation methods;simultaneously,due to the complex structure of the satellite power system and the large amount of telemetry data,data-based fault diagnosis methods are gradually replacing model-based methods have become a research hotspot.Based on this,this p aper takes geosynchronous satellite(GEO)power subsystem as the research object,carries out system-level mathematical modeling and fault simulation,and uses the data-driven fault diagnosis method based on deep learning to its typical fault mode for diagnosis.The research content of this paper mainly includes: Firstly,on the basis of mastering the working principle of GEO satellite power supply subsystem,choose the appropriate topology structure and determine the basic composition of each module(including solar battery array,storage battery,power controller).According to the set technical indicators of the simulation system,the simulation model of the satellite power system is built to achieve system-level simulation and analysis;Then,summarizing the typical failure mode of the satellite power supply system,realizing process controllable and repeatable fault simulation through the flexible fault injection method,the key parameters that reflect the state of the system are monitored and collected;Finally,Based on the obtained monitoring data,a fault diagnosis method of the satellite power supply system based on temporal convolutional network(TCN)is proposed,and the state of each time step during the operation of the system is recognized by a sequential classification method.By introducing the convolution architecture into the fault diagnosis field,it is directly oriented to multi-dimensional time series without manual feature extraction and selection to realize end-to-end fault diagnosis.Finally,the diagnosis results are evaluated and analyzed by using reasonable evaluation indexes,and compared with classical algorithms in the field of sequence modeling.The research results show that the satellite power system simulation model established in this paper conforms to the system indexes set by simulation and can correctly reflect the working state of GEO satellite power subsystem.The simulation results of typical fault modes of each part are matched with the operation mechanism of the system,which can provide support for subsequent fault diagnosis.The data-driven fault diagnosis method based on TCN is proposed to minimize manual involvement and realize "end-to-end" fault diagnosis.The results are evaluated through evaluation indicators.Compared with LSTM and Bi LSTM,the accuracy of the proposed method is improved while considering the operating efficiency,and the effectiveness of the diagnostic algorithm is verified.