Abstract: Time series is a set of data points or observed values arranged in chronological order, which is widely used in the fields of finance, meteorology and stock market analysis. Abnormalities in these data may mean potential problems, abnormal events or system failures. To facilitate further research on the design of anomaly detection methods for time series in the future, the related concepts of time series anomaly detection are introduced firstly. Secondly, the anomaly detection methods for univariate and multivariate time series at home and abroad are analyzed. After that, some general datasets of anomaly detection for time series are introduced and the performance of common methods on these datasets are compared. Finally, the key research directions on the design of anomaly detection method for time series in the future are discussed, which can provide a reference for relevant theoretical and applied research.

Key words: time series, anomaly detection, univariate time series, multivariate time series, general dataset