Abstract: Flight data is typical multivariate time series.It extracts the features of flight data based on SVD，gets candidate sets of similarity query by singular distance filtering，and constructs similarity measure modal via coordinate transformation theory in linear space，then realizes precise matching on candidate sets and gets ultimate results.At the same time it gives the strategy of conflict resolving for similar subseries，deeply analyzes the consistency and completeness of similarity query，points out that the methods in this paper will inevitably lead to false dismissals，and proposes a solution to decrease false dismissals and enhance query efficiency by merging experiential rules.At last the experiments on real data show the validity of the research.

Key words: multivariate time series, similarity query, flight data, feature extract, similarity measure

摘要： 飞行数据是一种典型的多元时间序列数据，基于奇异值分解提取飞行数据序列的特征，通过奇异值距离过滤获得相似模式查询的候选集，依据线性空间中的坐标变换原理构造多元时间序列的相似性度量模型，从而实现候选集上的精确匹配并获得最终的结果集。给出了相似子序列的冲突消解策略，深入分析了查询的完备性问题，指出该方法可能导致误判或成为误判的根源，提出融入先验规则来减少误判并提高查询效率的方法。在真实飞行数据上的实验结果验证了方法的有效性。

关键词: 多元时间序列, 相似性查询, 飞行数据, 特征提取, 相似性度量