Federated Learning Framework for Early Detection of Reconnaissance Attacks in Smart Grid Environments

Utility companies like industries, and manufacturing plants interact or transmit the requirements with consumers through the electrical grids. These electrical grids have been advanced using digital technology for controlling, securing and monitoring. The recent development of smart grids also raises the concern of being compromised by unauthorized users which leads to serious data breaches. The proposed research work in this paper is based on federated learning that helps the smart grid system to enhance its security. The model states the use of local trainings of multiple devices such as smart meters, communication networks, internet of things devices for safeguarding the privacy data and, Global training at every sub-station or the computing point to help the local devices address the malwares, cyber-attacks easily. The model can continuously improve its ability to detect attacks based on the real-time data collected from multiple IoT devices. This would enable the smart grid system to get resistant to cyber-attacks and failures. Federated Stochastic Gradient Descent (FSGD) model is proposed with reconnaissance attack dataset for enhancing the global learning of the model and that strengthens the privacy and trust on the proposed model.