Abstract

This master's thesis presents a comprehensive study on false data injection (FDI) attacks in photovoltaic (PV)-integrated power distribution systems (PDSs) and proposes two data-driven detection frameworks to identify such attacks against voltage regulation in both steady-state and transient modes. With the deployment of information and communication technologies (ICTs) for voltage regulation, PDSs are exposed to cyber threats, including FDI attacks.

The first proposed framework employs a supervised machine learning (ML) approach based on a support vector machine (SVM) to detect FDI attacks during the transmission of data from the centralized controller to PVs in a transient state. The framework collects a dataset of different operating points, such as loading conditions, to train the SVM. The performance of the trained framework for attack detection is compared with other supervised and unsupervised ML-based techniques. The results demonstrate the superior performance of the proposed framework in detecting FDI attacks against modified IEEE 33-bus PDS.

The second proposed framework is a convolutional neural network (CNN) approach to detect FDI attacks against voltage regulation in steady-state. The framework creates a comprehensive and realistic dataset that covers all normal conditions and unpredictable changes of a PDS during two years, including features such as voltage measurements, season, weekdays/weekends, load conditions, and PV generation power. The CNN is trained to distinguish normal grid changes from FDI attacks. The performance of the trained framework has been compared with other supervised ML-based and deep-learning techniques for FDI attacks against a modified IEEE 33-bus and 141-bus PDS to show the scalability of the proposed method, and the results demonstrate the superior performance of the proposed framework in detecting FDI attacks.

Overall, this study aims to enhance the cyber-security of PV-integrated PDSs by proposing two effective and reliable detection frameworks against FDI attacks in both steady-state and transient modes. By utilizing ML and deep learning techniques, the proposed frameworks are capable of accurately detecting FDI attacks, thereby improving the overall resilience of PDSs against cyber threats.