An effective parameter tuned deep belief network for detecting anomalous behavior in sensor-based cyber-physical systems

Cyber-physical system (CPS) integrates cyber systems and the corporal world to perform critical processes that are begun from the developments in digital electronics. The sensor placed in CPS is used to monitor and control processes that are vulnerable to anomalies (reliability failures and cyber-attacks). For the detection of anomalies and prevention of destructive issues, Abnormality Detection System (ADS) is introduced. But ADS suffered from a false alarm rate and missed detection rate, which has resulted in the degraded performance in CPS applications. A novel deep learning approach for detecting Anomalous Behavior in sensor-based CPS is presented in this study. Preprocessing is used in the conceptual scheme to remove noisy data. CPS irregularity is then detected using a DL-based deep belief network (DBN) framework. Furthermore, the number of hidden layers, number of iterations, period frequency, and learning rate of DBN are all tuned using a squirrel search algorithm (SSA). Several tests were conducted with the help of data from device CPS to validate the performance of the SSA-DBN model. To assure the SSA-DBN framework's appropriate detection accuracy, a complete comparable study is conducted, and the produced extensive experiments confirmed the framework's outstanding quality on the applicable information over the comparable approaches.