Research on Anomaly Monitoring Method for the Entire Chain of Minute-level Data Acquisition
Hui Li, Xingyuan Fan, Junyi Chen, Haitong Gu · 2025
In order to improve the detection timeliness of fullchain anomaly monitoring and reduce the missed detection rate, a new minute-level data acquisition full-chain anomaly monitoring method is proposed. Firstly, a multi-feature extraction model is constructed, integrating dimensional information such as time series volatility, statistical deviation, and cross-node correlation, and achieving precise aggregation of feature sub-vectors through the attention mechanism. Subsequently, a classification model is established based on the decision tree algorithm, and the feature similarity is calculated by recursively dividing the sample space. Ultimately, anomaly monitoring rules are formulated to achieve real-time anomaly determination throughout the entire chain of minute-level data collection. The entire process, from feature extraction to monitoring, forms a complete closed loop, taking into account both timeliness and accuracy. The experimental results show that in the anomaly detection of minute-level data acquisition, this method takes only $\mathbf{1 - 2}$ seconds, and the missed detection rate is stable at about $1 \%$, which is significantly superior to the comparison methods and performs excellently in both timeliness and accuracy.