In recent years, with the development of various industries, we have access to an increasing amount of data with timestamps, and people are increasingly valuing the industrial value brought by the temporal features within this data. Research on temporal features can extract patterns from historical data, which can then be utilized for the analysis and modeling of time series. Among the studies of time series, anomaly detection is an important branch. Detecting anomalies in time series enables the monitoring of data generated in real-time, thereby reducing data management costs, which is a meaningful practical problem. The industrial time series data nowadays exhibits characteristics such as high dimensionality and large volume, posing greater challenges for anomaly detection in time series. This is primarily manifested in traditional models performing reasonably well on low-dimensional and small-scale data but struggling to model a large number of high-dimensional temporal features. Additionally, anomaly definitions vary across different industrial scenarios, and data labels are often limited. Consequently, one of the challenges faced by anomaly detection in high-dimensional time series is how to achieve good detection results in various scenarios with limited data labels, where samples are restricted. Therefore, this paper proposes a Transformer model combined with data augmentation methods to investigate algorithms for anomaly detection in small-sample time series.