一种不确定连续时间序列的Top-K异常检测算法

针对噪声数据对时间序列异常检测准确性的影响问题, 提出了一种不确定连续时间序列Top-K异常检测算法。在典型时间序列异常检测方法的基础上对时间序列的异常值进行区间处理, 构造满足均匀分布的密度函数, 结合不确定Top-K技术, 实现含噪连续时间序列在分布未知情况下的Top-K异常排序。实验部分采用模拟数据和真实数据进行算法测试, 算法较传统方法在异常检测的准确率方面有明显提高, 虽然在计算时间上有所增加, 但提出了相应的优化策略, 使计算时间在k值大于5时有明显改善, 验证了算法的有效性。

Uncertain continuous time series Top-K anomaly detection method

Aimed at the problem that the noise data influence on the anomaly detection results for time series, this paper put forward a kind of uncertain continuous time series Top-K anomaly detection algorithm. Based on the typical time series anomaly detection method, it dealed with time series discord as interval, and structured density functions of uniform distribution. Then it combined with the uncertain scores Top-K technology to implement Top-K ranking on uncertain time series data with noise and unknown distribution. Experiment tests on simulated data and real data, this proposed algorithm has increased significantly in anomaly detection accuracy than traditional time series detection method. Although the method has increased in computing time, this paper put forward the corresponding optimization strategy and improved on computing time obviously when the K value was large. It verifies the effectiveness of the algorithm.