Intrusion Detection Algorithm of CAN Bus Data Flow Based on Inception-ResNet Model

With the rapid development of the industrial internet, the volume of data in industrial networks is increasing, and network attacks are becoming more concealed and complex. Traditional machine learning methods are no longer suitable for the new scenarios of industrial intrusion detection. Therefore, this study addresses the difficulties in feature selection and overfitting when applying traditional machine learning methods to handle industrial bus data. It proposes a Deep Convolutional Neural Network Intrusion Detection (DCNN-IDS) model and verifies it using the natural gas dataset from Mississippi State University. The DCNN-IDS model utilizes the Inception-ResNet model as its underlying structure, optimizes the model size, redesigns the network architecture, and introduces an image generator module to generate two-dimensional CAN data images with sequentially ordered bit identifiers. Experimental results demonstrate that the DCNN-IDS model defeats other machine learning algorithms and exhibits excellent performance in detecting intrusion attacks in industrial networks.