摘要
模式挖掘是应用于卫星智能监控服务中的一项重要技术.当前频繁模式挖掘的使用率要远远高于稀有模式挖掘,然而对于卫星遥测数据流来说,频繁模式挖掘在安全监测和故障预防等方面所取得的成效不如稀有模式挖掘.因为频繁模式挖掘无法从卫星的遥测数据中揭示卫星可能存在的潜在故障.卫星遥测是持续不断进行的,所以其数据流存在数据量大、传输速度快和数据重复性高的特点.如果采用一般的稀有模式挖掘方法来挖掘卫星数据流,尽管其速度比频繁模式挖掘快,但总体上仍然较慢,不能满足卫星实时监测的需要.针对上述问题,本文提出一种可快速找出卫星遥测数据流中隐藏信息的最小稀有模式挖掘方法,它具有如下优点:(1)无需卫星领域知识;(2)引用滑动窗口技术并将主观参数(窗口尺寸)客观化,使得算法能够实时地处理数据流;(3)通过仅挖掘最小稀有模式方式来提高算法的挖掘效率;(4)该算法使用双向遍历技术提高算法的运行速度.从某在轨卫星的遥测数据流中选取10个关键特征参数进行算法验证.实验结果表明,本文所提算法能有效地从卫星遥测数据流中挖掘出全部的最小稀有模式,并且其挖掘速度比现有的方法快.
The pattern mining is an important technology applied to satellite intelligent monitoring services. At present, the usage ration of frequent pattern mining is much higher than that of the rare pattern mining. However, as far as the research whose target object is satellite telemetry data stream, the results of the frequent pattern mining is not as effective as those of the rare pattern mining in terms of the security monitoring, fault prevention and so on. This is due to the fact that the frequent pattern mining can not reveal the potential faults and hidden dangers that may exist in the components of satellites from satellite telemetry data stream, and it cannot also provide preventive measures. Since telemetry for satellites is continuously carried out, the data stream on satellite telemetry has the following characteristics: in the first place, it includes a large amount of data;in another, its transmission speed is fast, and what is more, its data repetition rate is high. If the common rare pattern mining methods are used to mine satellite telemetry data stream, although its mining speed is faster than the mining speed of the frequent pattern mining, it is still slow on the whole and cannot meet the needs of real-time monitoring for satellite. What is more, the general rare pattern mining mainly takes static data as the object of the research. In order to find out all the rare patterns, the strategy of generating all candidates and then pruning is usually adopted, but which may be leading to the problem of combinatorial explosion and the speed of algorithm is slower. With the development of the big data and the technologies of cloud platform, traditional storage devices have been unable to keep up with the speed of data transmission. Therefore, if static data is still used as the object of the research, the storage device may easily lose data when storing the data. In order to solve the above problems, this paper proposes a minimal rare pattern mining method that can quickly discover the hidden information in the satellite telemetry data stream. It has the following advantages: First, the algorithm proposed in this paper, that is, the minimal rare pattern mining algorithm, does not require knowledge in the satellite field. Second, this paper references the sliding window technology and objectifies the subjective parameters (that is, window size) so that the algorithm proposed in this paper can process satellite telemetry data stream in real time. Third, the algorithm improves the mining efficiency by only mining the minimal rare pattern, instead of all the rare patterns. Fourth, the algorithm uses bidirectional traversal techniques to improve the speed of the algorithm. Ten key features are selected from the satellite telemetry data stream of some orbiting satellite to verify the proposed method in this paper. The experimental results indicate that the algorithm proposed in this paper can effectively mine all the minimal rare patterns from satellite telemetry data stream, and the speed of the proposed algorithm is faster than that of the existing methods.
作者
周忠玉
皮德常
ZHOU Zhong-Yu;PI De-Chang(College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106)
出处
《计算机学报》
EI
CSCD
北大核心
2019年第6期1351-1366,共16页
Chinese Journal of Computers
基金
国家自然科学基金(U1433116)
中央高校基本科研业务费专项资金(NP2017208)
南京航空航天大学研究生创新基地(实验室)开放基金(kfjj20181605)资助~~
关键词
最小稀有模式
卫星
遥测数据流
自顶向下
自底向上
双向遍历
模式挖掘
数据挖掘
minimal rare pattern
satellite
telemetry data stream
top-down
bottom-up
bidirectional traversal
pattern mining
data mining
