Abnormal Data Detection of Engineering Archives Based on Transformer and Density Clustering

Power enterprise engineering archives data holds significant value but is plagued by quality issues, such as missing or anomalous data. Traditional abnormal data detection methods perform poorly because they overlook the systemic characteristics and complex correlations within the data. This paper proposes an abnormal data detection method for engineering archives, which is based on the Transformer model and density clustering. Utilizing the Transformer's powerful feature extraction and longrange semantic understanding capabilities, the method extracts complex correlation features from engineering archives data. This provides more effective feature inputs for density clustering, enhancing clustering accuracy for anomaly detection, followed by data cleaning to ensure correctness. The performance of rule-based, clustering-only, and the proposed method were compared. The results demonstrate that the proposed method significantly outperforms the others, better adapting to the complex tasks of engineering archives abnormal data detection.