基于LLM的时间序列异常子序列检测算法

Outlier subsequence detection algorithm for time series based on LLM

为了提高时间序列中异常子序列检测算法的有效性,提出一种基于局部线性映射(LocalLinear Mapping,LLM)的异常子序列检测算法.该算法将时间序列子序列通过其相邻子序列线性重构,很好地保留了时间序列子序列与其相邻子序列的相关性.基于LLM的映射特性,使用两种异常指标(贡献因子,重构误差),并将其应用于ST东方(B)股票交易时间序列数据集的异常子序列检测中.实验结果表明,所提出的算法对异常子序列的异常检测具有很好的效果,有效提高了时间序列中异常子序列的检测效率.

In order to improve the effectiveness of outlier subsequence detection algorithm for time series, the detection algorithm for outlier subsequence based on Local Linear Mapping （LLM） was presented. In the present algorithm, the subsequence in time series was mapped through the linear reconstruction of its neighbors, and the relativity between the subsequence in time series and its neighbors was well preserved. Two outlier indices including the contribution factor and reconstruction error were applied to the outlier subsequence detection process for the ST Orient （B） stock time series data sets based on the mapping characteristics of LLM. The experimental results show that the proposed algorithm is effective in the outlier detection of outlier subsequence and can improve the effectiveness of the outlier subsequence detection.