Association Mining Of Satellite Parameters Based On Machine Learning

Ground Support System is an important service platform for the space science missions of The Strategic Priority Program on Space Science,such as the Hard X-ray Modulation Telescope(HXMT).A main task for the GSS is to analyze large amount of parameter data and monitor the platform and on-board equipment during the in-orbit operation.It is an important method and basis for post-analysis to excavate the correlation relation between satellite parameters and extract the objective relationship and regulations contained in the parameters,which has high practical application value.This thesis analyses and researches the existing correlation analysis algorithms,and designs several satellite parameter association mining algorithms based on LightGBM machine learning model according to the characteristics of satellite parameter data.The experimental verification of HXMT satellite in-orbit data shows that the designed algorithm can extract highly correlated parameter groups from the mass parameters.According to the characteristics of the large number and complex value of satellite parameters,this thesis firstly divides HXMT satellite parameters into binary classification parameters,multi-classification parameters and regression parameters according to the number of types of parameter values.Secondly,according to the number of parameters involved in the analysis,the satellite association mining algorithm is divided into satellite single-parameter association mining algorithm and satellite multi-parameter association mining algorithm.In this thesis,a satellite single-parameter association mining algorithm based on LightGBM machine learning model is designed.This algorithm can quantify the importance of features used in LightGBM modeling,and can quickly evaluate the degree of correlation between massive parameters,which is suitable for the satellite binary classification parameters.In this thesis,real in-orbit parameter data of HXMT are used for experimental verification.The results show that when the target parameter is highly correlated with a single parameter,this parameter combination can be found out successfully,but it is also found that more parameters are strongly correlated with multiple parameters.This thesis designs two kinds of satellite multi-parameter association mining algorithms,which are suitable for all kinds of satellite parameters.Among them,according to the total number of satellite parameters involved in the analysis,the preliminary filtering algorithm dynamically adjusts the length of the interval describing the number of parameters increased in each iteration.The algorithm is flexible,has a low complexity and a simple idea,in order to quickly narrow the search scope of high related parameters.In order to improve the accuracy of recognition,the advanced filtering algorithm only chooses the most relevant parameters in the current set of parameters in each iteration.After several iterations,all the high-correlation parameters are finally obtained.As a cost,the advanced filtering algorithm has high complexity.In this thesis,real in-orbit parameter data of HXMT are used for experimental verification.The results show that for binary parameters with only two values of parameters,all parameters highly related to target parameters can be accurately mined.For multi-classification parameters and regression parameters with more complex and changeable values,the evaluation algorithm can preliminarily screen out the parameter set with higher correlation.Combined with the characteristics of satellite parameters,this thesis independently designed the parameter association mining algorithm.Through experimental verification,the highly correlated parameter sets of the target parameters concerned can be mined from the massive parameters.The research results of this thesis can be used in the real-time monitoring system of in-orbit status of space science missions.This indicates that the association mining algorithm based on machine learning model proposed in this thesis has high practical application value and provides important reference opinions for the post analysis of in-orbit parameter data of space science missions.