Enhancing Trust Modelling for the Internet of Underwater Things

Abstract

The Internet of Underwater Things (IoUT) has gained growing interest from researchers and industry alike, due to its potential for advancing the development of smart cities and underwater intelligent systems. However, the harsh and unpredictable nature of underwater environments, coupled with the inherent limitations of existing technologies, presents significant challenges to establishing a sustainable IoUT. Furthermore, the open nature of such networks renders them highly susceptible to malicious attacks and security threats. Traditional security measures, which are widely implemented in conventional cyber systems, exhibit severe performance constraints in underwater networks, highlighting the urgent need for novel security solutions that meet the unique requirements of underwater networks. Trust modelling has been widely recognised as an effective soft security measure to mitigate the impact of internal attacks. It primarily achieves this by analysing behavioural characteristics between network entities, thereby introducing a layer of defence against malicious activities. In the context of underwater networks, trust establishment between nodes has the potential to significantly enhance overall network security. However, existing Trust Modelling and Management (TMM) often fail to address the complexities of underwater environments, which necessitate new TMM that are lightweight, accurate, and decentralised. In light of these limitations, this thesis investigates and enhances TMM to meet the application requirements of underwater networks while addressing the specific challenges inherent to IoUT. The central research question addressed in this thesis is: To what extent can existing TMM accommodate diverse network topologies within the IoUT and effectively mitigate potential attacks from both the communication and physical domains. In order to answer this question, a comprehensive understanding of the key challenges and potential application requirements for underwater networks is required. To facilitate this investigation, a simulated environment is constructed to analyse the effectiveness of TMM. This study critically evaluates the capabilities of current TMM in detecting malicious activities across various underwater network structures, identifying vulnerabilities, and exposing potential attack vectors. In response to these findings, this thesis proposes a distributed multi-dimensional TMM, referred to as the Mobility-Aware Trust Model (MATMU), designed to enhance the detection of malicious behaviour within the constraints of underwater environments. MATMU expands the metric domain to include mobility-aware metrics, allowing for the assessment of similarities and differences in node movement patterns. Additionally, the model employs a dynamic weighting strategy that integrates metrics from both the communication and physical domains. The performance of MATMU is evaluated through extensive simulations conducted across various underwater scenarios and attack models. The results demonstrate that MATMU effectively mitigates malicious behaviour, exhibiting notable improvements over benchmark models, particularly in terms of faster convergence and enhanced attack detection. These findings underscore the suitability of MATMU for strengthening secure and reliable communication in underwater networks. This thesis also tackles the critical issue of dishonest recommendations within TMM in the IoUT context, which is introduced by malicious entities, aiming to manipulate trust computations by providing false or misleading recommendations, thereby degrading the reliability and stability of the TMM. A novel recommendation evaluation method is introduced, combining filtering and weighting strategies to more effectively detect dishonest recommendations. The proposed model incorporates an outlier detection-based filtering technique and deviation analysis to evaluate recommendations based on both collective outcomes and individual experiences. Furthermore, a belief function is employed to refine recommendations by assigning weights based on criteria such as freshness, similarity, trustworthiness, and trust decay over time. This multi-dimensional approach demonstrates a marked improvement in recommendation evaluation, effectively capturing deceptive behaviours that exploit the complexities of IoUT. The effectiveness of the model is validated through extensive simulations and comparative analyses with existing trust evaluation methods, demonstrating consistently high performance across varying proportions of dishonest recommendations, with the highest accuracy improvement observed when dishonest recommendations constitute up to 45% of the total recommendations. These findings underscore the model’s potential to significantly enhance the reliability and security of IoUT networks.