Research On Data Mining For Spacecraft Telemetry Time Series Data

The spacecraft as the main carrier of the aerospace industry,is the foundation of human space exploration.Maintaining the regular work of spacecraft is vital to the implementation of the entire aerospace industry,it should be the priority of the research work.Compared with ground simulation,spacecraft telemetry data can reflect the real working state and reliability of spacecraft.It is the main bases for spacecraft performance monitoring and state analysis.Use relevant data and extract the information effectively,not only provided support for the management and decision-making of spacecraft,but also increased the reference value for spacecraft improvement.This paper is based on nearly 2 million telemetry data of a large spacecraft from 2011 to 2015.And according to the characteristics of telemetry parameters,this paper discussed the design and implement time series sequence of feature representation,furthermore,similarity measurement and center calculation the three aspects of the algorithm are demonstrated as well.In this paper,the main research and innovation points as follows:(1)Focusing on the characteristics like rich parameters and complex types of spacecraft telemetry data,GIE-APLA algorithm which adaptive piecewise linear representation method was designed based on global information entropy.The method makes up the deficiency of PLA method in calculating efficiency,and uses information entropy to measure the fluctuation of current data segment in order to achieve the purpose of adaptive division in linear time.The linear regression is used to fit the original sequence to ensure the accuracy of feature representation.(2)A new dynamic time warping algorithm ASDTW based on adaptive line segments was proposed to overcome the shortcomings of existing time series similarity measurement methods.In order to solve the problem of excessive computation cost of DTW algorithm,firstly,GIE-APLA algorithm was used to express the original sequence as a sequence segment.And according to the distance between the defined sequence of the geometric characteristics within the period,the sequence segment as a basic unit to improve the traditional point by point match strategy caused by excessive computational overhead in the dynamic matching stage.The experimental results show that the ASDTW algorithm under the premise of guarantee matching accuracy,solved the problem of the DTW algorithm computational overhead.(3)In order to overcome the shortcomings of large computation overhead and sensitive to merge sequence,a central sequence algorithm based on sequence segment SSB is proposed.Firstly,in order to reduce the influence between different forms of sequence,the similarity division was conducted to the sequence set through hierarchical clustering;then solve the center sequence in each sequence subsets with the method of iteration.Considering the dynamic and iterative computation overhead caused by matching,in each iterative process,using sequence matching to reduce the amount of calculation and reduce the impact of the merge order by defining the centroid sequence.The experimental results showed that the SSB derived center sequence is better than NLAAF in characterization ability,its performance is better than DBA,and it is superior to the above two algorithms in computation efficiency.