Implementation of adaptive efficient CapNet for identifying cyber anomalies and attacks in cyber-physical systems using Stacked Autoencoder-based deep feature engineering

Cyber-physical systems (CPS) are described as the latest invention of complicated networks, which combine both the physical and cyber worlds. Because of their strong coupling nature, it is considered that a few elements may fail to be recognized within a particular duration. Deep learning models offer effective approximate actions and representation-based learning abilities, permitting conventional techniques to understand the most effective guidelines for sequential decision-making within complicated surroundings, particularly while handling high-dimensional inputs and huge-scale issues. Therefore, a new strategy is recommended to identify cyber anomalies and attacks within the CPS using adaptive deep learning. Initially, the required input data are fetched from various benchmark datasets, and these gathered data are fed into the feature‑extraction procedure. Here, the Stacked Autoencoder is utilized for extracting the deep features. Further, the resultant features are given as input to the Adaptive Efficient Capsule Network to determine cyber anomalies and attacks in the CPS. Here, a Modified Random Number-based Garter Snake Optimization Algorithm is used for optimally tuning the parameters in the CapsNet. Finally, detailed experiments are carried out for the developed cyber‑anomaly and attack‑detection approach to guarantee the effectiveness of the implemented mechanism over traditional techniques.