False data injection attack is one of the important threats affecting the safe operation of power grids

and existing detection methods rely on power grid topology for spatial feature extraction

limiting their scalability. Moreover

the deep spatial information in power grid measurement data is not fully extracted

which limits the localization and detection performance. Therefore

this paper proposes a false data injection attack location detection method based on image coding and multi-head self-attention convolutional neural network. In this method

the one-dimensional historical cross-sectional power grid data is encoded by principal component analysis and Gramm’s angle field to extract the deep spatial information in the power grid measurement data

which overcomes the limitations of the existing detection methods that rely on the topology of the power grid for spatial feature extraction. On this basis

the multi-head self-attention mechanism is introduced into the two-dimensional convolutional neural network to pay more attention to the important features with rich deep spatial information in the two-dimensional image data. The effectiveness of the proposed detection method is verified through simulations on IEEE-14/57 node systems

with comparative analysis indicating superior performance in precision

accuracy

recall

and F1-score across different scenarios.