Abstract:To realize the classification of illegal actions in intelligent monitoring grid operations, improve the efficiency and safety of power system operation and maintenance, and reduce the dependence on sample labeling, this paper proposes a method which is based on cross-domain few-shot learning for illegal action classification in power grid operations. In this method, we designed an innovative cross-domain alignment mechanism which helps the classification model to better understand and adapt to the feature changes between different domains, and enhances the model"s invariance learning ability in the target domain, thereby improving the accuracy and efficiency of the classification of illegal actions under different objective factors in the power grid operation scenario. Experimental results show that the effectiveness of the proposed method in reducing the time cost of sample annotation and the dependence on large-scale annotated data, and through the cross-domain few-shot learning method, the annotated source domain data and the unlabeled target domain data are input into the classification model, which realizes the efficient and accurate classification of illegal actions in the power grid operation scenario, and shows a wide application prospect in the actual power grid operation and maintenance.