Classification of System Failures in Cyber-Physical Systems Using Heterogeneous Edge Enhanced Graph Attention Networks

From the past few years, Cyber-Physical Systems (CPS) have rapidly evolved in various sectors by integrating computing, networking, and physical processes. CPS facilitates the management of critical infrastructure such as transportation, healthcare, and energy. Even though there are advanced technologies, which are vulnerable to failures led to severe consequences including unpredictable interactions between cyber and physical components. Therefore, to address these issues a robust model is proposed by using Heterogeneous Edge-Enhanced Graph Attention Network (HEE-GAT). Initially, the data is collected from BoT-IoT database which consists of huge traffic data. Then, the collected data is preprocessed using Z-score normalization is applied to standardize feature distributions while preserving network traffic characteristics. Furthermore, for selection of significant features Skill Optimization Algorithm (SOA) is utilized. Finally, HEE-GAT is utilized for effective classification of system failures in CPS. Unlike existing failure detection models such as LSTM-GNN hybrids, which struggle with capturing heterogeneous dependencies and dynamic edge relationships in CPS data, HEE-GAT enhances graph-based learning by assigning dynamic edge weights and utilizing attention mechanisms to distinguish between different failure types with higher accuracy. The proposed HEE-GAT model acquired accuracy of 95% when compared to the existing Fractional Artificial Protozoa Optimization enabled Spiking VGG-16 (FAPO Spiking VGG-16).