The temporality and redundancy of electrical power measurement data pose significant challenges to the accuracy and speed of detecting false data injection attacks. This paper proposes a localization detection algorithm based on a temporal autoencoder and a correlation-labeling convolutional neural network to achieve rapid and precise location detection of false data injection attacks. The algorithm uses the temporal autoencoder to reconstruct temporal measurement data

highlighting anomalous features with error data to enhance the algorithm's precision. Additionally

by introducing the correlation label enhancement method

the one-dimensional convolutional neural network can utilize the correlations between data to reduce data redundancy

achieving rapid and accurate attack location detection. The algorithm's effectiveness is verified through simulation experiments on the IEEE-14 and IEEE-57 bus test systems

and it demonstrates significant advantages in terms of precise localization detection and comprehensive generalization capability compared to existing machine learning algorithms. To enhance the interpretability of the algorithm

principal component analysis is used for the visualization analysis of data

showing the process of the algorithm's handling of anomalous information.