IoV Intrusion Detection Using Improved CNN for Lower False Alarms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intrusion detection methods for Internet of Vehicles (IoV) are inefficient and inaccurate due to the complex network topology and diverse communication protocols, leading to prolonged detection times and high false alarm rates.
Innovation Solution
An improved convolutional neural network-based method for IoV intrusion detection that includes data preprocessing, convolutional calculations, pooling operations, and end-to-end classification using a SoftMax layer, with backpropagation for parameter optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms (SVM, DNN, MLP) are used for intrusion detection, then detection accuracy can be improved, but training time and detection time increase significantly
Solution Approach 1:
The patent segments the intrusion detection task into two distinct phases: an offline training phase using machine learning algorithms to build detection models, and an online detection phase using these pre-trained models for real-time traffic analysis. This segmentation allows complex training to occur when time is not critical, while real-time detection uses lightweight models for speed.
Solution Approach 2:
The patent performs preliminary action by pre-training detection models offline using historical traffic data and attack patterns. These pre-trained models are then deployed for real-time detection, eliminating the need for time-consuming training during actual intrusion detection operations.
2Reliability
If deep learning models are trained on large datasets to improve detection accuracy, then false alarm rate decreases, but computational resources and training time increase
Solution Approach 1:
The patent divides the computational workload into offline model training (using large datasets to reduce false alarms) and online detection (using pre-trained models with minimal computation). This segmentation allows extensive training to occur when computational resources are available, while real-time operation consumes minimal energy.
Solution Approach 2:
The patent creates copies of pre-trained detection models that can be deployed across multiple IoV nodes. Once a model is trained with extensive computational resources, identical copies can be distributed and executed efficiently on resource-constrained devices without requiring the same level of computational power.
3Ease of operation
If traditional intrusion detection methods are applied to IoV environment, then detection capability is provided, but detection accuracy is insufficient due to complex network topology and diverse protocols
Solution Approach 1:
The patent transforms intrusion detection from a protocol-specific task to a pattern recognition task by changing the parameters from protocol fields to traffic flow features. This parameter transformation enables the use of machine learning models that can generalize across diverse IoV protocols and complex network topologies without requiring protocol-specific detection logic.
Data Source
AI summary
The present disclosure belongs to the technical field of Internet of vehicles (IoV) security and provides an IoV intrusion detection method and device based on an improved convolutional neural network. The method of the present disclosure includes: collecting original data of data traffic during IoV communication, and inputting the original data to a data dimension reduction algorithm model for IoV intrusion detection for preprocessing to obtain standardized data for IoV data analysis; inputting the standardized data for IoV data analysis to an improved convolutional neural network model for calculation, including: performing convolutional calculation and nonlinear activation on the input data for layering; performing two convolutional operations, two pooling operations and one full connection operation on each layer of data; and classifying a data set output by the improved convolutional neural network model through a SoftMax layer.