The smart grid is a multi-dimensional datagenerating cyber-physical system. Distributed architectures
and the heterogeneous nature of the Internet-of-Things (IoT)
sensors make it more prone to various cyber-attacks. False
data injection attacks (FDIAs) have recently emerged as
significant threats to smart grid state estimation. As a result,
real-time locational detection of stealthy FDIAs is critical for
smart grid security and reliability. In this paper, we introduce
a comparative analysis of various deep-learning approaches to
test their effectiveness in the location-based detection of FDIA.
Also, a deep learning approach is developed by constructing
a multi-feature architecture based on a convolution neural
network and long short-term memory network (MCNNLSTM). Extensive testing on IEEE test cases has demonstrated
that the proposed approach outperforms the existing deep
learning approaches in locating FDIAs for small and large
systems under different attack scenarios. We evaluate the
performance of each model in terms of presence and locationbased detection accuracy, model complexity, and prediction
time. Extensive results in the IEEE 14 and IEEE 118-bus
systems show that the suggested architecture has a locational
detection accuracy of more than 94% and 95%, respectively.
From the results, we can conclude the proposed approach is
more robust, scalable, and faster in detecting the locations
of compromised measurements than the other deep learning
models.