一种基于LLM的高维时间序列数据异常检测方法

以国家重大建设项目稽察中的数据一致性判别问题为应用背景，针对时间序列型高维数据提出了一种基于局部线性映射(Local Linear Mapping，LLM)的数据变换方法，该方法将各高维数据点通过其相邻点的线性重构映射至低维空间，从而很好地保留了高维空间中各数据点与相邻数据点的相关性。基于LLM的映射特性，提出了三种异常指标，并将其应用于面向国家重大建设项目稽察数据一致性判别问题的高维时间序列数据异常检测中。数值计算表明，所提出的方法对时间序列异常检测具有很好的效果，适合于较大规模高维时间序列数据的异常检测应用。

An Outlier Detection Method in High Dimensional Time Series Based on LLM

Outlier detection in large high dimensional data sets is an important research direction in data mining.A new data transformation technique called Local Linear Mapping (LLM) is first proposed in which each data point in time series is mapped into low dimensional vector space through the linear reconstruction by its neighbors, and then based on the properties of LLM, three outlier indices are presented and applied to the outlier detection process for high dimensional time series data sets. Numerical computation shows that the presented approach is effective in detecting outliers in high dimensional time series data sets.