The automated operations of Industrial Control Systems (ICSs) highly rely on the interconnected devices, sensors, and actuators that are monitored and controlled by the Supervisory Control and Data Acquisition (SCADA) systems. Despite the numerous benefits of unifying the networking technologies with SCADA systems, ICSs are more susceptible to cyber-attacks that can disrupt the secure operations of the Critical Infrastructures (CIs). Thus, the design and development of an efficient attack detection approach has become a complex task. Hence, this research work presents a novel hypergraph based anomaly detection technique with Enhanced Principal Component Analysis and Convolution Neural Network (EPCA-HG-CNN) to detect deviant behaviours of such systems. The proposed EPCA-HG-CNN algorithm involves two phases: (i) Dimensionality reduction using enhanced PCA and (ii) Anomaly detection with hypergraph based Convolution Neural Network. Further, the performance of EPCA-HG-CNN is evaluated with SUTD's Secure Water Treatment System (SWaT) and the experimental results show that the proposed EPCA-HG-CNN has identified anomalous behavior of the data with high detection rate, low false positives, and better classification accuracy.

The most effective method for insulation assessment in electrical power apparatus is Partial Discharge (PD) detection. During measurements the interference from background environment hampers the PD signal and reduces its measurement accuracy. This paper discuss on the implementation of a Hankel matrix based Fast Singular Value Decomposition (H-FSVD) technique for removing noise from the PD signals. The data is first represented into Hankel Matrix (HM) structure, with appropriate sampling then using Lanczo process the Hankel matrix size is reduced and through Singular Spectral Analysis thresholds are fixed for noise detection and removal. This algorithm has been examined on simulated as well as PD signals measured on two different laboratory environments from transformers with real and simulated noise.The experiment is part of a series of experiments to detect PD patterns related to realistic transformer defects. The denoising performance of H-FSVD is compared with the wavelet based denoising methods, Empirical Mode Decomposition method and normal SVD. Numerical results show that H-FSVD efficiently removes the noise with less computation time, even for large size data.