Towards Smart and Secure Industry 4.0: Hybrid Optimization and Transformer Driven Security Framework for Cyber Threat Detection

The necessity to identify the cyberattack in the systems of the Industrial Internet of Things (IIoT) has become topical since the latter is a key element of the Industry 4.0 services such as smart manufacturing, energy control, and industrial communications. The rapid growth in the number of connected devices and cyber-physical objects has exposed IIoT environments to sophisticated cyber-attacks, including DDoS attacks, malware, and zero-day attacks potentially leading to disruption of operations, data integrity and data safety. The existing intrusion detection systems have severe problems with high complexity of calculation, low ability to adapt to novel attacks, and low detection. To overcome these issues, this paper will propose a hybrid cybersecurity solution that will use the Dandelion Optimization Algorithm (DOA) to select the features in addition to a new Longformer Vision Randomized Network (LoV-RaNet) to classify the attack. The selected features are most relevant and discriminative as DOA reduces redundancy hence lowering the computational cost whereas LoV-RaNet utilizes long-range attention and randomized residual connections to accurately identify known and unknown attacks. The proposed framework enables the scalable, efficient and real-time IIoT intrusion detection. The experimental results also show that LoV-RaNet is more effective, with the accuracy of 0.990, the precision of 0.985, the recall of 0.982, and F1-score of 0.988 being higher than the conventional intrusion identification models.