Since the existing line loss data anomaly identification method cannot determine the cause of abnormal line loss and the low recall rate, an anomaly identification method of massive daily line loss data of power grid based on multi-dimensional features is proposed. Firstly, integrating all kinds of power operation data, a multi-dimensional feature daily line loss data anomaly traceability model is established, and the Pearson correlation coefficient is used to calculate the correlation between transformer and voltage between different lines, so as to clarify the cause of line loss anomaly. Then, we normalize the daily line loss data, introduce the concept of time dispersion to evaluate the abnormal degree of line loss, obtain the abnormal calculation model of line loss data under load variation by using neural network learning, and input the load values of different nodes to complete the abnormal identification operation of daily line loss data. The simulation results show that the proposed method can judge the cause of abnormal line loss, and the recall rate of abnormal identification is as high as 94%, which improves the accuracy of abnormal identification of massive daily line loss data, and solves the deviation problem of abnormal calculation of line loss data.