چكيده به لاتين
Advances in computing and networking has provided new possibilities for physical systems that could not be feasibly added before. This has led to the emergence of engineering systems which called cyber-physical systems (CPS), which defined as collaborative environments consisting of computational and communicational elements controlling physical entities with the help of sensors and actuators. A CPS integrates physical devices (i.e., sensors, actuators) with cyber components (i.e., networks, servers). Pervasive healthcare systems, smart grids, and unmanned aircraft systems are examples of CPSs that have become highly integrated in the modern world. As this integration deepens, the importance of securing these systems increases. Therefore, the detection of abnormalities in these systems is very important.
In these systems, intrusion detection systems (IDSs) analyze each data sample independently (behavior-based intrusion detection, knowledge-based intrusion detection, anomaly-based intrusion detection and etc.) and ignore the main characteristics of data samples (i.e. their sequential nature). It is obvious that most abnormalities can be detected only by analyzing a sequence of data samples and will not be detectable by IDSs due to the independent analysis of each sample. Since the nature of time series depends on their observations and the sequence of observations, time series analysis can be a good solution for anomaly detection in a sequence of data.
The purpose of this thesis is to propose a method for anomaly detection in CPSs through analyzing sensor data. The proposed method receives data from the sensors first, register them as time series, model the training datasets, predicts the future behavior of the system, and finally, model the test datasets, then abnormalities are detected by comparing the predicted behavior. Experimental results show that this method has high true detection rate and low false one.
