A4 Vertaisarvioitu artikkeli konferenssijulkaisussa
Lightweight AI-Driven Real-Time Intruder Detection System for IoT-Enabled Smart Meters
Tekijät: Ben Dhaou, Imed; Mihoub, Alaeddine; Krichen, Moez
Toimittaja: Kallel, Slim; Raibulet, Claudia; Bouassida Rodriguez, Ismael; Faci, Noura; Bennaceur, Amel; Cheikhrouhou, Saoussen; Ayed, Leila Ben; Sellami, Mohamed; Nakagawa, Elisa Yumi; Halima, Riadh Ben
Konferenssin vakiintunut nimi: International Conference on Service-Oriented Computing
Kustantaja: Springer Nature Singapore
Julkaisuvuosi: 2025
Lehti: Lecture Notes in Computer Science
Kokoomateoksen nimi: Service-Oriented Computing – ICSOC 2024 Workshops : ASOCA, AI-PA, WESOACS, GAISS, LAIS, AI on Edge, RTSEMS, SQS, SOCAISA, SOC4AI and Satellite Events, Tunis, Tunisia, December 3–6, 2024, Revised Selected Papers, Part I
Vuosikerta: 15833
Aloitussivu: 251
Lopetussivu: 262
ISBN: 978-981-96-7237-0
eISBN: 978-981-96-7238-7
ISSN: 0302-9743
eISSN: 1611-3349
DOI: https://doi.org/10.1007/978-981-96-7238-7_20
Julkaisun avoimuus kirjaamishetkellä: Ei avoimesti saatavilla
Julkaisukanavan avoimuus : Ei avoin julkaisukanava
Verkko-osoite: https://doi.org/10.1007/978-981-96-7238-7_20
Smart meters are essential to the advancement of the smart grid, enhancing energy generation, distribution, and consumption through advanced functionalities. These devices facilitate two-way communication, demand-response initiatives, real-time usage monitoring, and dynamic pricing, thereby optimizing decision-making and energy efficiency for both providers and consumers. However, the security of smart meters is paramount, as data breaches can lead to user attacks, privacy violations, and manipulations of the energy market. To address these vulnerabilities, this paper presents a Lightweight AI-based Intrusion Detection System (IDS) designed for real-time security in smart meters, leveraging TensorFlow for its core operations. Our research includes a thorough comparative analysis of various lightweight classification algorithms to identify the most effective approach for detecting intrusions. We further categorize and examine five distinct types of energy theft, utilizing a public dataset to strengthen the robustness and applicability of the IDS. Optimized for embedded systems, the proposed IDS is rigorously tested on a Raspberry Pi, demonstrating its viability for resource-constrained devices in smart metering. Built on a two-layer artificial neural network (ANN), the system excels at recognizing normal and anomalous patterns, achieving an accuracy of 97.08% and an F-score of 97.11%, all while maintaining low CPU usage at just 8 s. This research significantly enhances the security and reliability of smart grid systems, fosters trust, and encourages the adoption of sustainable energy solutions. Furthermore, the successful deployment of the Raspberry Pi highlights the practicality of our Lightweight AI-driven IDS, establishing a benchmark for high-efficiency, low-resource intrusion detection in critical infrastructure components.
