Rare Pattern Mining And Application On Telemetry Time Series Data

Satellite telemetry data is typical time series data,which contains a large number of objective laws.Rare pattern is an uncommon data pattern that is usually used for knowledge analysis in expert fields.It can reveal hidden rare rules and abnormal information from telemetry time series data.In this paper,the telemetry time series data of a satellite is taken as the research object,and in-depth research is performed on the minimal rare pattern mining and the perfectly rare pattern mining,and it is applied to the field of satellite outlier detection.Telemetry time series data has the characteristics of large data volume,high dimension and high repeatability.Aiming at the problems that the existing rare pattern mining methods are difficult to efficiently analyze satellite large data streams,and frequent patterns cannot reflect rare rules,a minimal rare pattern mining method that can quickly find hidden information in satellite telemetry data streams is proposed.This method has the following advantages: First,it does not require knowledge in the satellite field.Second,referencing sliding window technology and objectiveizing the subjective parameter(window size).Third,improving the mining efficiency of the algorithm by mining only the minimal rare pattern.Fourth,using bidirectional traversal technology improves the running speed of algorithms.Experimental results indicate that this method can effectively mine all the minimal rare patterns from the satellite telemetry data streams,and its mining speed is faster than the existing methods.The perfectly rare pattern is a rare pattern that contains only rare items.It plays a very important role in expert fields such as adverse drug reaction monitoring,disease prediction of rare symptoms,and satellite outlier detection.However,the existing perfectly rare pattern mining methods have the problems of low time efficiency and inability to mine large telemetry data streams.In view of the above problems,two new algorithms are proposed,namely a perfectly rare pattern mining method based on the BitSet structure and a perfectly rare pattern mining method based on the improved FP-tree structure.The former needs to scan the database once and the latter needs to scan the database twice without generating a candidate set.Experimental results show that these two algorithms can efficiently mine perfectly rare patterns from large satellite telemetry data streams and can be applied to other fields.Outlier detection is a common method for analyzing satellite data streams which aims at ensuring the stable operation of satellites and extending the life of the satellites.In the existing outlier detection methods,most of methods compute distance of points to solve certain specific outlier detection problems.However,for the outlier detection of satellite data streams,there are still many issues with these methods.Due to the large data volume and high dimensions of satellite data streams,the computational cost of these methods is high,and satellite data streams cannot be processed quickly,and all outliers cannot be found.In addition,most of the existing methods are based on static datasets or common data streams,they cannot meet the processing requirements of satellite telemetry data.In order to be able to efficiently process satellite telemetry data streams,an outlier detection method based on rare patterns is proposed.This method uses the minimal rare pattern as the input subspace and determines the outliers in the telemetry data by calculating the value of the new outlier identification factor.Experimental results show that the method can be applied to outlier mining in the satellite field,and is superior to the existing methods.