Abstract: In order to accurately detect abnormal voltage and current data in the distribution network and solve the problem of low accuracy of the detection model caused by the scarcity of abnormal data under normal operation of the distribution network, a method for detecting abnormal data based on an improved chaos optimization algorithm (ICEO) - dual attention mechanism Transformer (DAM Transformer) is proposed. This method first utilizes the strength controlled diffusion anomaly synthesis (SDAS) method to generate partial anomaly data, in order to alleviate the problem of insufficient model recognition accuracy caused by the scarcity of real anomaly samples; Secondly, an innovative DAM Transformer model was proposed, which integrates a dual attention mechanism to achieve collaborative modeling of complex patterns in different time scales and feature spaces, effectively improving the identification of multi-scale feature coupling relationships in the context of abnormal distribution network data; Finally, ICEO was used to iteratively optimize the hyperparameters of DAM Transformer, further improving the optimization efficiency and generalization performance of the model in complex scenarios. The results show that compared with traditional models, this method improves the accuracy of identifying abnormal voltage in distribution networks by 12.81% and the accuracy of identifying abnormal current by 12.22%. In data scarcity scenarios, the recognition accuracy is significantly better than traditional models. This method effectively solves the core bottleneck of sample scarcity and difficulty in modeling multi-scale features in abnormal data recognition of distribution networks, improves the accuracy of abnormal recognition and the stability of model operation, and provides key technical support for digital inspection, real-time fault warning, and operation and maintenance decision optimization of intelligent distribution networks. It has engineering application prospects.