Study On Similarity Search Algorithm Based On QAR Flight Time Series Data

To solve the rapid growth of aircraft maintenance workload, limited troubleshooting time and maintains of low efficiency contradictions, current domestic airlines generally use QAR for the aircraft flight data records. QAR records every moment of the true state of the flight and provides a strong basis by similarity search for equipment fault diagnosis.In this thesis, QAR flight data which is used for fault diagnosis is studied. The users can get acquainted with the status of airplane in time and well informed by the fault information after analyzing the flight data. The users achieve fault location based on time series similarity search and improve the level of fault diagnosis.The main achievements in this thesis:First this thesis analysis the structure of QAR flight data parameters and classifies the similar parameters; According to the process of data mining, this thesis implements data preprocessing of QAR flight data, using linear regression and k-nearest neighbor algorithm two ways to fill vacancies values; Using moving averages, moving window least square method polynomial smoothing and roughness penalty smoothing three kinds of smoothing method to smooth QAR flight data and compare the results; Using DFT, DWT, PAA, SAX to transform the QAR flight data. After comparing of different methods, SAX symbolic is better.During describing, storing and retrieving such operation, SAX can not overcome time series amplitude flex and timeline drift, so improved algorithm is proposed. QAR flight data will be divided into three stages and use algorithms to fill the cruise stage, thus allowing effective search for time series of different length. This thesis does similarity search through creating index by the fault model. Given the index tree, similarity sequence can be quickly and efficiently retrieved with the query sequence. The experiment and item prove the feasibility and practical significance of improving plane troubleshooting.