| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.26 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A Internet das Coisas (IoT) tem revolucionado o setor tecnológico em diversas aplicações, tais como cidades inteligentes, medidores de temperatura e casas inteligentes, permitindo a objetos enviar informação online em tempo real. Long-Range Wide Area Networks (LoRaWAN) é um protocolo IoT que se destaca pela sua capacidade de comunicação a longa distância mantendo um baixo consumo de energia, através da sua arquitetura de rede aberta e distribuída. No entanto, apesar de este protocolo estar a ser cada vez mais adotado por organizações, empresas e outras entidades, o mesmo apresenta vulnerabilidades de segurança que, apesar de muitas vezes negligenciadas, podem ser exploradas por meio de ataques informáticos que comprometem os serviços fornecidos por quem usufrui da tecnologia ao violar a privacidade e integridade dos dados. Com isso, foi proposto o desenvolvimento e implementação de um sistema de deteção de intrusões para redes LoRaWAN, que analisa o tráfego de rede LoRaWAN e classifica os pacotes como apresentando um comportamento anómalo ou não anómalo, com base nos padrões de comportamento aprendidos por algoritmos de Machine Learning implementados. Através de um estudo inicial do protocolo LoRaWAN e dos seus mecanismos de segurança, chegou-se à conclusão de que, apesar desses mecanismos protegerem o protocolo de certos ataques, os mesmos não são capazes de o proteger
de outros ataques que apresentam um grande risco para a segurança do protocolo, tais como Jamming. Também foi efetuado um estudo teórico de algoritmos de Machine Learning e Deep Learning adequados para detetar ataques não resolvidos pelo protocolo LoRaWAN. O sistema de deteção de intrusões foi avaliado usando um conjunto de dados de pacotes LoRaWAN de um gateway de Lisboa (2020), com anomalias sintéticas injetadas para teste. Resultados mostraram que Histogram-Based Outlier Score alcançou os melhores resultados, embora a validação em ataques reais permaneça como trabalho futuro.
Abstract Internet of Things (IoT) has been innovating the technology sector in applications such as smart cities, temperature gauges and smart homes, allowing objects to send information online in real time. Long-Range Wide Area Networks (LoRaWAN) is an IoT protocol that stands out for its ability to communicate over long distances while maintaining low energy consumption, through its open and distributed network architecture. However, even though this protocol is being increasingly more adopted by organizations, companies and other entities, it presents several security vulnerabilities which, despite being many times neglected, can be exploited through cyberattacks that disrupt the services provided by those who use the technology, violating data privacy and integrity. Considering these aspects, it was proposed the development and implementation of an intrusion detection system for LoRaWAN networks, which analyses LoRaWAN network traffic and classifies packets as either presenting an abnormal or normal behaviour, based on behavioural patterns learnt by implemented Machine Learning algorithms. An initial study of LoRaWAN and its security mechanisms showed that, although these mechanisms protect the protocol from certain attacks, these mechanisms can’t protect the protocol from another attacks that present a great risk to the protocol security, such as Jamming. A theoretical study of Machine Learning and Deep Learning algorithms that can be adequate for detecting attacks not resolved by LoRaWAN also took place. The intrusion detection system was evaluated using a dataset of LoRaWAN packets from a Lisbon gateway (2020), with synthetic anomalies injected for testing. Results showed that Histogram-Based Outlier Score achieved best performance, though validation on real attacks remains future work.
Abstract Internet of Things (IoT) has been innovating the technology sector in applications such as smart cities, temperature gauges and smart homes, allowing objects to send information online in real time. Long-Range Wide Area Networks (LoRaWAN) is an IoT protocol that stands out for its ability to communicate over long distances while maintaining low energy consumption, through its open and distributed network architecture. However, even though this protocol is being increasingly more adopted by organizations, companies and other entities, it presents several security vulnerabilities which, despite being many times neglected, can be exploited through cyberattacks that disrupt the services provided by those who use the technology, violating data privacy and integrity. Considering these aspects, it was proposed the development and implementation of an intrusion detection system for LoRaWAN networks, which analyses LoRaWAN network traffic and classifies packets as either presenting an abnormal or normal behaviour, based on behavioural patterns learnt by implemented Machine Learning algorithms. An initial study of LoRaWAN and its security mechanisms showed that, although these mechanisms protect the protocol from certain attacks, these mechanisms can’t protect the protocol from another attacks that present a great risk to the protocol security, such as Jamming. A theoretical study of Machine Learning and Deep Learning algorithms that can be adequate for detecting attacks not resolved by LoRaWAN also took place. The intrusion detection system was evaluated using a dataset of LoRaWAN packets from a Lisbon gateway (2020), with synthetic anomalies injected for testing. Results showed that Histogram-Based Outlier Score achieved best performance, though validation on real attacks remains future work.
Descrição
Palavras-chave
IoT LoRaWAN Sistema de deteção de intrusões Machine learning Deep learning Intrusion detection systems