اميد رستمي

Detecting anomalies and attacks in cyber-physical systems (CPS) are of paramount importance owing to their growing prominence in controlling critical systems. However, this is a challenging task due to the heterogeneity and variety of components of a CPS, and the complex relationships between sensed values and potential attacks or anomalies. Such complex relationships are results of physical constraints and domain norms which exist in many CPS domains. One of the approaches to secure cyber-physical systems is the addition of advanced security features. Considering various constraints such as physical environment feedback, distributed control, real-time responsiveness, wide-scale geographic distribution, and multi-layered features, it is impossible to implement complex security system. The required power and cost also impose limitations on the design of cyber-physical systems, making it challenging to incorporate advanced software and hardware on them. In this thesis, a method has been proposed for detecting anomalies in cyber-physical systems. In the first step, the data is cleansed for the identification and correction or removal of inconsistencies. In this method, to address the issue of model overfitting and data balancing, the oversampling approach has been employed. In the next step, the PCA approach has been employed for data visualization and dimensionality reduction to two dimensions. Also, an autoencoder has been used to improve the performance of machine learning models, engineering, selection and extraction of useful features. In the anomaly detection stage, a random forest has been employed. For eva‎luation, the SWaT dataset, which has been published by the iTrust researchers, has been used. The eva‎luation results of the proposed method demonstrate a true positive rate (TPR) of 1.0 and a false positive rate (FPR) of 0.0003 The proposed method exhibits a significantly better false positive rate and a false negative rate close to zero compared to similar works.

سيستم‌هاي سايبر-فيزيكي , كشف ناهنجاري , كدكننده خودكار , جنگل تصادفي , يادگيري ماشين

Cyber-Physical Systems , Anomaly Detection , Auto Encoder , Random Forest , Machine Learning

Omid Rostami

Mohammad Abdollahi Azgomi