One of the objectives of D-HYDROFLEX is to facilitate the digitalization of hydropower plants. As a part of this digitalization initiative, it is essential to ensure data privacy and cyber-resilience measures. The F-BOX tool developed by MINDS undertakes this role, by leveraging deep learning techniques and AI models trained in a federated fashion to enhance intrusion and cyber-threat detection.
In the context of D-HYDROFLEX, F-BOX will serve as a mechanism to detect intrusions and anomalies on the industrial communication protocols used in a hydro power plant’s infrastructure. To accomplish this, AI models will be trained with the goal to successfully detect potential intrusion on the networking infrastructure. To ensure data privacy, the training of the corresponding models is conducted via federated learning (FL). According to FL, participating clients interested in training AI models, can cooperatively build a shared model in a decentralized fashion via the aid of a central server. This eliminates the need of transferring raw data to the server, and thus, preserving privacy.
Apart from AI-based intrusion detection capabilities, F-BOX also supports explainable AI (xAI) functionalities, helping security analysts to understand certain outputs of the generated security events, e.g., why an observed activity in the industrial control system of the hydropower plant is malicious or benign. Additionally, F-BOX leverages custom FL aggregation strategies, tailored to tackle the challenges imposed by heterogeneous data among participating clients.
