In view of low checking efficiency of abnormal electricity data in some power grid companies

in this paper a kind of dual-layer checking method based on stacking strategy and deep learning is proposed. Firstly

the mixed sampling technique is adopted for comprehensive preprocessing of input data

which effectively solves such issues as data missing and class imbalance in large-scale power data. Then

multiple shallow machine learning models are used as base-classifiers and the stacking strategy is used for fusion so to fully leverage the advantages of each base classifier. After that

the deep neural networks are adopted as meta-classifier to find the optimal fusion rules and out-put the final results

which effectively improves the checking accuracy. Finally

the test is performed on the real electricity dataset

and the the effectiveness of the proposed method is proved by the result.