Next–Generation Intrusion Detection for Iot evcs: Integrating cnn, lstm, and gru models

Download 6 Mb. Pdf ko'rish
bet	6/16
Sana	18.02.2024
Hajmi	6 Mb.
	#158491

1 2 3 4 5 6 7 8 9 ... 16

Bog'liq
mathematics-12-00571

Figure 2.
Ensemble Model Design and Architecture.
The proposed model begins with an input layer designed to receive a vectorized form
of network traffic data. The input shape is tailored to the dimensions of the feature set
extracted from the network packets. The data undergo a series of transformations through
two Conv1D layers, each followed by a MaxPooling1D layer to reduce dimensionality and
emphasize the most salient features.
The sequential aspect of the data is then processed through a hybrid LSTM-GRU
arrangement-a single LSTM layer with return sequences set to true feeds into a GRU layer,
creating a deep, sequential model capable of handling complex time-dependent patterns.
This combination captures a comprehensive temporal profile of traffic data, encompassing
both short-term fluctuations and long-term trends indicative of intrusive behavior.
To finalize the classification, the model is flattened and passed through a dense layer
with ReLU activation, introducing nonlinearity and aiding in learning complex patterns.
A dropout layer is included to mitigate overfitting, followed by a softmax activation layer
that classifies the traffic data into predefined categories ranging from normal to various
types of attack vectors.
The model is compiled with the Adam optimizer, which is known for its efficiency
in handling large datasets and its adaptive learning rate capabilities. The loss function
employed is sparse categorical cross-entropy, which is particularly suited for classification
problems in which the classes are mutually exclusive.
This innovative NIDS framework is designed to be a cornerstone in defense against
cyber threats in IoT-based EVCS. Using a DL approach that integrates CNN layers for
feature extraction with LSTM and GRU layers for temporal data analysis, the proposed
model not only identifies existing threat patterns but also adapts to emerging anomalies.
It stands as a testament to the potential of AI in fortifying the cybersecurity measures of
critical infrastructure within the smart city ecosystem.
4.3. Integration of CNN, LSTM, and GRU
In the quest to bolster the cybersecurity infrastructure of IoT ecosystems, particularly
EVCS, the integration of CNN, LSTM, and GRU presents a formidable approach [
30
].
This confluence of neural network models forms a sophisticated analytical framework
capable of discerning intricate patterns within sequential data that are characteristic of
intrusion activities.

Mathematics 2024, 12, 571
11 of 26
The synthesis of CNN, LSTM, and GRU models facilitates a dual-phase analysis–spatial
followed by temporal–each phase tackling different data aspects. CNNs are adept at ex-
tracting spatial features from input data (Figure
3
). They apply a series of learnable filters
to input sequences, capturing essential features such as the signature of packet headers or
the frequency of specific network events. This is particularly crucial in the context of EVCS,
where the myriad of devices and communication protocols introduce a complex and dense
feature space [
12
].

Download 6 Mb.

1 2 3 4 5 6 7 8 9 ... 16

Download 6 Mb.

Pdf ko'rish