Bayesian methods for modelling and management of trust in wireless sensor networks
- Publication Type:
- Issue Date:
Security and trust are two interdependent concepts and are often used interchangeably when defining a secure wireless sensor network (WSN) system. However, security is different from trust in that, it assumes no node is trustworthy and requires ongoing authentication using sophisticated protocols leading to high communication and computation overheads. This makes the traditional cryptographic security tools hard, if not impossible, to be used in wireless sensor networks that are severely resource constrained. Trust on the other hand is the exact opposite of security in that any node can interact with any other and requires no authentication and unwrapping of hidden keys to carry on with their business and hence carries zero overhead. However, this leads to the miss-use and abuse of networks causing loss and damage to the owners of the networks. This thesis focuses on developing novel methods for modelling and managing trust that enable WSN to be secure while significantly reducing computing and communication overheads. Although researchers have been studying the problem of trust modelling and management in wireless sensor networks for over a decade, their focus was on the trust associated with routing messages between nodes (communication trust). However, wireless sensor networks are mainly deployed to sense the world and report data, both continuous and discrete. However, there are no methods in the literature that focus on the trust associated with misreporting data (data trust). In this thesis, we model the trust associated with the integrity of the data, and propose methods to combine the data trust with the communication trust to infer the total trust. Bayesian probabilistic approach is used to model and manage trust. A new risk assessment algorithm for establishing trust in wireless sensor networks based on the quality of services characteristics of sensor nodes, using the traditional weighting approach is introduced. Then a Beta distribution is used to model communication trust (due to its binary nature) and determine the weights in terms of the Beta distribution parameters to probabilistically combine direct and indirect trust. The thesis extends the Bayesian probabilistic approach to model data trust for cases when the sensed data is continuous. It introduces the Gaussian trust and reputation system to that accounts for uncertain characteristics of sensor data. Finally we introduce a Bayesian fusion algorithm to combine the data trust and communication trust to infer the overall trust between nodes. Simulation results are presented to demonstrate how the models accurately classify different nodes as being trustworthy or not based on their reliability in sensor reporting and routing functions.
Please use this identifier to cite or link to this item: