Trust Evaluation Method Based On Trusted Computing In Edge Computing

In recent years,the number of Internet of things access terminals has increased dramatically,which makes the traditional cloud computing insufficient to support the storage and process of massive derived data.As a possible computing paradigm of IoT,edge computing has the characteristics of providing point-to-point services and helps to share computing tasks in the cloud,so it has been widely concerned and studied.However,multi-level collaboration in the edge computing environment involves resource request and access as well as data transmission.Resources usually change dynamically and data types are different from each other.The distributed characteristics of edge computing make it very difficult to establish a trust mechanism similar to cloud computing.Therefore,considering the demand of edge computing for real-time security services,this thesis focuses on the reasonable edge computing architecture and the corresponding trust evaluation method,which provides a reasonable basis for the subsequent task selection and allocation,resource scheduling and optimization.This thesis proposes an edge computing architecture based on application services and resource collaboration scenarios,defines the concepts involved,and proposes the corresponding trust evaluation method based on the definitions.In this method,three key attributes,i.e.identity,behavior and capability,are selected to evaluate the trust of the edge device.The comprehensive trust attribute of the device is expressed as a feature vector.Then the related attributes are quantified and the corresponding calculation method is given,and the calculation complexity of this method is discussed.Next,this thesis proposes a trust prediction algorithm based on capsule neural network(TPCN),which takes the trust feature vector of the device as the input of the network,and then predicts the overall trust value in the current environment.Using the characteristics of shallow layer and short training process,the algorithm can quickly predict the overall trust value in a small-scale edge computing environment.Simulation results show that the prediction algorithm has good performance in global convergence time,malicious device detection rate and task failure rate.According to the dynamic and distributed characteristics of large-scale edge computing environment,this thesis proposes a distributed dynamic trust evaluation method.The trust evaluation of terminal usually involves two processes: the selection of terminal to edge server and the trust evaluation of edge server to terminal.This method mainly considers the factorsthat affect the trust evaluation from the above two perspectives.The calculation method given can adjust the trust weight according to the dynamic changes of the environment.After that,a random gradient descent algorithm based on fully connected neural network(FCNSGD)is proposed to predict the trust value of the whole environment.The simulation results show that the algorithm has better performance in global convergence time,malicious device detection rate and task failure rate in a large-scale distributed environment,but it needs a certain time model to achieve faster performance compared with the general algorithm This problem is the main direction of later research.The two algorithms proposed in this thesis are used to solve the trust evaluation problem of edge computing,TPCN is suitable for small-scale centralized network environment,FCNSGD is suitable for large-scale distributed network environment.Experimental results show that the two algorithms have good performance in dealing with their own problems.