Smart Meter Modbus RS-485 Intrusion Detection by Federated Learning Approach

Abstract

To accelerate digital transformation and assemble "connecting the world," the IoT has been invented. To make and made more convenient in our daily life activities, billions of devices have been connected so far. Recently, we have noticed how they are helping cyber-physical systems (CPSs) to reach more elevated levels of evolution. Amidst diverse CPSs entities, the smart grid, Advanced Metering Infrastructure (AMI), is among the foremost essential entities since its rapid transformations. Wherein the Modbus RS-485 protocol is typically used in smart meters for physical layer communication. The key concern resides in fact that an attacker may easily compromise the smart meter systems since it lacks authentication and encryption mechanisms. As a countermeasure, an intrusion detection system (IDS), by applying a Federated Learning (FL) approach, could be an effective solution to detect the malicious activities of the RS-485 communication network, ensuring data protection from intruders. Since its built-in data protection mechanism and model could train without sharing sensitive private data. Henceforth, this research proposes a Federated Learning-based IDS for detecting critical attacks against the smart meter. We experiment with Modbus Attack DataSet for AMI (MAMI) datasets, and experiment results depict that the FL approach is reasonably effective in detecting critical smart meter attacks, moreover, protects the data privacy concern. The Multilayer Perceptron (MLP) classifier outperforms, which achieves a detection accuracy and detection rate of 99.98%, respectively.