محمد لشكري بخشايش

Today, traditional power grids are no longer sufficient to meet users’ needs an‎d often face short comings in areas such as consumption management, energy optimization, an‎d cybersecurity. These grids also encounter environmental challenges, including increasing carbon emissions. With ad vancements in electricity an‎d computing, smart power grids have been introduced to address users’ requirements an‎d establish two-way communication between electricity distributors an‎d consumers. This technology improves consumption management an‎d offers advanced security solutions, en hancing user trust. However, due to the complexity of smart power grids, these systems face new security threats. One fundamental challenge is maintaining security an‎d combating cyber threats, which directly impact human well-being. Therefore, securing these grids is crucial. One effec tive approach is the use of machine learning an‎d deep learning technologies. In this thesis, a novel methodbasedonfederatedlearningispresentedtodetectanomaliesan‎dattacksinsmartpowergrids. Data aggregators are used as federated learning nodes, an‎d the electricity distribution company acts as the central server. This approach maintains data privacy while ensuring high security an‎d accu racy in identifying cyber threats. To eva‎luate this method, two datasets, NSL-KDD an‎d UNSW NB15, were used, an‎d three deep learning models—FNN, RNN, an‎d 1D-CNN—were tested. The proposed model achieved 92% to 94% accuracy in identifying most anomalies. Analyses demon strated the method’s scalability an‎d strong performance in various conditions. This research lays a foundation for advanced machine learning approaches an‎d broader adoption of federated learning in smart power grids.

شبكه هوشمند برق , امنيت سايبري , حريم خصوصي , يادگيري مشاركتي , يادگيري ماشين

Smart Grid , cyber security , Privacy , federated learning , Machine Learning

Mohamad Lashkari

Dr. Mozayeni