In order to improve the efficiency and accuracy of electricity market settlement, an anomaly detection method is proposed for power settlement electricity data based on graph theory and hybrid convolutional neural network. Firstly, the input data is preprocessed using a hybrid resampling technique to solve the class imbalance problem in the data sets. Secondly, based on graph theory, the electricity data is transformed from a one-dimensional sequence structure to a two-dimensional graph structure, and the periodicity and temporal correlation characteristics of the graph structure are mined through graph convolutional network and convolutional neural network to improve the detection accuracy of abnormal electricity. Furthermore, a spatial attention mechanism is introduced into convolutional neural network to improve the detection performance of the model. Finally, abnormal data detection is performed on the actual power data sets. And the results show that the proposed method is superior in comprehensive performance such as accuracy and area under curve (AUC) value.