IMPROVING DETECTION OF FAULTS AND CYBERSECURITY IN POWER SYSTEM USING DIGITAL TWIN AND SVM

Modern power systems are increasingly susceptible to cyber-physical threats due to the widespread integration of renewable energy resources, digital monitoring, and automation technologies. Among these threats, False Data Injection (FDI) attacks pose a particularly serious challenge, as they can stealthily manipulate system measurements to mimic physical faults—potentially triggering inappropriate control responses and compromising grid stability. Machine learning (ML) classifiers, such as Support Vector Machines (SVM), often struggle to distinguish these attacks from real faults due to overlapping data characteristics and the absence of real-time contextual awareness. To address this critical challenge, this research proposes a novel hybrid framework that integrates a real-time Digital Twin (DT) with an SVM classifier to improve fault and cyberattack differentiation in power distribution systems. The DT is built on a modeled IEEE 33-bus system using MATLAB/Simulink and continuously generates dynamic, high-fidelity synthetic data that simulates normal, faulted, and cyber-compromised grid states. These DT-derived features—including residual deviations and rate-of-change indicators—serve as context-aware inputs to enhance SVM classification accuracy beyond what is possible using static measurements alone. This thesis fills a clear research gap by demonstrating that combining DT-generated dynamic data with conventional SVM learning allows for more accurate, scalable, and adaptive real-time anomaly classification. Performance evaluation based on voltage profile tracking, confusion matrices, and standard classification metrics (precision, recall, F1-score) reveals that the proposed DT-SVM model significantly outperforms the baseline SVM model. Classification accuracy improved from 78% (static SVM) to over 94.5% (DT-SVM). This work not only presents a computationally efficient and interpretable solution for distinguishing between cyberattacks and faults but also establishes a scalable and deployable methodology for improving cyber-physical resilience in smart grids. By leveraging real-time simulation and adaptive ML, the proposed framework offers a next-generation approach for intelligent monitoring and decision-making in secure power system operations.