CN-122021716-A - Space-time correlation driven edge node social trust evaluation method and system

CN122021716ACN 122021716 ACN122021716 ACN 122021716ACN-122021716-A

Abstract

The invention discloses a space-time correlation driven edge node social trust evaluation method and a space-time correlation driven edge node social trust evaluation system, and relates to the field of intersection of Internet of things technology and artificial intelligence. According to the method, the space-time behavior perception, trust dynamic quantification and strategy self-learning optimization integrated technical framework is constructed by introducing the space-time diagram attention network, the multi-scale trust evolution mechanism, the deep reinforcement learning and other leading-edge technologies, so that accurate evaluation and dynamic enhancement of node social trust in an edge ubiquitous intelligent scene are realized, the limitation of the traditional method in the aspects of dynamic property, real-time property and cross-domain adaptability is broken through, and the integral trust level of the network and the resource cooperation efficiency are remarkably improved.

Inventors

YANG BORAN
LIU JINGJING
Ye Binqiang
ZOU WEI
CAO KEXIN

Assignees

重庆理工大学

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (9)

1. A space-time correlation driven edge node social trust evaluation method is characterized by at least comprising the following steps: The method comprises the steps of S1, generating node behavior space-time embedded representation, acquiring historical interaction records, resource state data and network topology data of nodes in an edge ubiquitous intelligent network, and constructing a space-time dynamic graph; S2, multi-scale dynamic trust evaluation, namely inputting the node embedded representation sequence into a mixed model, wherein the mixed model is composed of a space-time diagram convolution network and a double-flow converter; Extracting local space-time characteristics of node behaviors through the space-time diagram convolution network; Through a normal trust evolution mechanism of the main stream learning node of the double-stream transducer on a long time sequence, and through a potential malicious behavior mode of the auxiliary stream sensing node; post-fusing the double-flow converter output to obtain a trust evolution state vector of the node; Based on the trust evolution state vector, respectively calculating short-term trust components, medium-term trust components and long-term trust components of the nodes by utilizing a multi-scale trust evolution mechanism, and carrying out weighted fusion through dynamic weight coefficients to output comprehensive trust values of the nodes; and S3, self-adaptive trust enhancement, wherein the comprehensive trust value and the current network state are used as a state space and are input into a trust enhancement strategy model based on deep reinforcement learning, the model outputs trust enhancement actions, and the actions are executed to optimize the overall trust level of the network.
2. The method for evaluating social trust of a space-time correlation driven edge node according to claim 1, wherein the step of generating an embedded representation of the node in the space-time diagram attention network in S1 is realized by the following formula: Wherein, the An embedded representation of the representation node i at time step t; A neighbor set of the node i; The associated weights of nodes i and j at time step t are calculated through an attention mechanism; Is a weight matrix which can be learned; Is a time decay coefficient; is a nonlinear activation function.
3. The method for evaluating social trust of a space-time correlation driven edge node of claim 1, wherein the space-time graph convolution network adopts a sandwich depth network structure with alternately stacked space graph convolution and time convolution; The space diagram convolution is used for extracting space characteristics related to network topology; the time convolution adopts a time sequence cavity convolution network to capture time sequence dynamic change.
4. The method for evaluating social trust of a space-time correlation driven edge node according to claim 1, wherein the post-fusion strategy for dual-stream transducer output is as follows: splicing the normal trust evolution feature vector output by the main stream transducer with the malicious behavior perception feature vector output by the auxiliary stream transducer; and then performing dimension reduction and fusion through a full connection layer to generate a final trust evolution state vector.
5. The method for evaluating social trust of a spatiotemporal association driven edge node according to claim 1, wherein the multi-scale trust evolution mechanism calculates the comprehensive trust value of the node by the following formula: Wherein, the Representing the comprehensive trust value of node i; 、 And Representing short, medium and long-term trust components, respectively; 、 And Is the dynamic weight coefficient of each component and meets + + =1。
6. A spatiotemporal correlation driven edge node social trust assessment method according to claim 5, wherein said short term trust component Calculating based on the behavior fluctuation rate of the node in the last time window; The behavior fluctuation rate is obtained by calculating the variance of the task success rate and the change rate of the resource contribution degree in the window; the mid-term trust component Based on the interactive trend statistical analysis of the nodes in the past days, adopting a linear regression model to fit the slope of the task completion quality of the nodes as a trend index; the long-term trust component Based on the historical performance of the node in the whole life cycle, the method obtains the weighted average success rate of all the historical interactions of the node by calculating, and introduces a time attenuation factor to enable the recent interactions to have higher weight.
7. The method for evaluating social trust of a space-time correlation driven edge node according to claim 1, wherein the trust enhancement policy model based on deep reinforcement learning is a deep Q network, and a state space s of the deep Q network is defined as: Wherein, T is the comprehensive trust value vector of all nodes in the network, R is the resource state vector of all nodes, L is the network link quality matrix, Q is the task queue state of each node; The action space a is a discrete operation set for raising, maintaining or lowering the trust value of a designated node, and the reward function r is designed as follows: Wherein, the Representing the change amount of the overall trust level of the network after the action is executed; representing the cost of system resources consumed to perform the trust-enhancing action; Penalty coefficients are cost.
8. A space-time correlation driven edge node social trust evaluation system is used for the space-time correlation driven edge node social trust evaluation method according to any one of the claims 1-7, and is characterized by comprising a data acquisition module, a behavior modeling module, a trust evaluation module and a trust enhancement module; The data acquisition module is used for periodically or based on event triggering acquiring historical interaction records, task completion quality, resource contribution degree and network topology parameters of the nodes from the edge ubiquitous intelligent network; the behavior modeling module is connected with the data acquisition module, receives the output of the data acquisition module based on a space-time diagram attention network architecture, and generates node embedding through the space-time diagram attention network; the trust evaluation module is connected with the behavior modeling module and is used for receiving the space-time embedded representation of the node, and calculating and outputting the comprehensive trust value of the node by utilizing a space-time diagram convolution network, a double-flow Transformer and a multi-scale trust evolution mechanism; The trust enhancement module is connected with the trust evaluation module, receives the comprehensive trust value and the current network state, and generates and coordinates and executes an optimal trust enhancement strategy by adopting a trust enhancement strategy model based on deep reinforcement learning.
9. The spatiotemporal correlation driven edge node social trust evaluation system of claim 8, wherein the trust evaluation module comprises a hybrid model sub-module, a multi-scale evaluator sub-module, and a trust output sub-module; the mixed model submodule consists of a space-time diagram convolution network and a double-flow converter and is used for extracting and fusing characteristics of an input node embedded sequence and outputting a trust evolution state vector; the multi-scale evaluator sub-module is connected with the mixed model sub-module and is used for respectively calculating short-time, medium-term and long-term trust components according to the trust evolution state vector and carrying out weighted fusion; The trust output sub-module is connected with the multi-scale evaluator sub-module and is used for outputting a final node comprehensive trust value and transmitting the final node comprehensive trust value to the trust enhancement module.

Description

Space-time correlation driven edge node social trust evaluation method and system Technical Field The invention relates to the crossing field of the Internet of things technology and artificial intelligence, in particular to a space-time correlation driven edge node social trust evaluation method and a space-time correlation driven edge node social trust evaluation system. Background With the rapid development of the Internet of things and artificial intelligence technology, edge intelligence is gradually evolving into an edge ubiquitous intelligent paradigm. Currently, edge deployment of computing power resources is deepened continuously, and network architectures with cloud edge end cooperation are widely applied in a plurality of fields. Particularly in high-computation-power demand scenes such as metauniverse, video generation, automatic driving and the like, the traditional edge server is difficult to meet the increasing computation demands due to fixed deployment places and limited single equipment resources. This current situation motivates a paradigm shift from edge intelligence to edge ubiquitous intelligence. The edge ubiquitous intelligent fills the gap of the demand of computing resources by fully utilizing the communication, calculation and storage resources of mass social idle intelligent devices (such as intelligent network-connected automobiles, home large model gateways, holographic game workstations and the like) at the edge of the network, and creates great extendability economic growth potential for the efficient utilization of the social idle resources. However, in a highly open, dynamic mobile environment, which is edge ubiquitous intelligence, the social trust relationship between nodes is significantly affected by the spatiotemporal dynamics of multi-dimensional resource collaboration. The resource state, interaction behavior pattern and network topology of the nodes all evolve continuously with time and space, which makes trust a complex, dynamically evolving function instead of a static attribute. At present, researchers at home and abroad conduct related researches on a social trust evaluation and management method. DinI.U., banoA., awanK.A. et al in the "LightTrust:LightweighttrustmanagementforedgedevicesinIndustrialInternetofThings"【IEEEInternetofThingsJournal,2023,10(4):2776-2783】 article propose a lightweight trust management scheme based on historical interactions that enables terminal nodes to communicate without performing complex trust calculations. According to the method, the calculation cost is reduced, but only static historical interaction records are considered, and the node resource state and the space-time dynamic change of the network topology cannot be perceived, so that the assessment accuracy is insufficient in a highly open edge ubiquitous intelligent environment. LiuY., haoX., renW. et al propose a trust management scheme oriented to the zero-trust internet of things in "Ablockchain-baseddecentralized,fairandauthenticatedinformationsharingschemeinzerotrustInternetofThings"【IEEETransactionsonComputers,2023,72(2):501-512】 article, and ensure the reliability of shared information and fairness to participating nodes through a blockchain technology. The scheme improves the safety of the system, but a dependent global consensus mechanism introduces significant communication and delay expenditure, and is difficult to meet the harsh requirement on trust evaluation instantaneity in an edge ubiquitous intelligent scene. LimI.S., masudaN. in "Totrustornottotrust:evolutionarydynamicsofanasymmetricn-playertrustgame"【IEEETransactionsonEvolutionaryComputation,2024,28(1):117-131】 paper, an asymmetric multi-party trust game evolution model is designed based on the evolution game theory, sequential interaction and trust losing behaviors of participants are considered, and positive influence of an incentive mechanism on trust is verified. The model reveals a game mechanism of trust evolution, but focuses on theoretical analysis, is not combined with real-time resource cooperative behavior of nodes, and lacks the capability of accurate and quantitative evaluation in an actual dynamic network environment. WangQ., zhaoW., yangJ. et al in "C-DeepTrust:Acontext-awaredeeptrustpredictionmodelinonlinesocialnetworks"【IEEETransactionsonNeuralNetworksandLearningSystems,2023,34(6):2767-2780】 propose a context-aware deep trust prediction model that calculates the context trust relationship between nodes by cosine similarity of feature vectors. The model effectively supports information interaction decisions of the online social network, but the 'context' mainly surrounds social attributes, and ignores space-time contexts formed by general computing and memory resource collaboration in edge ubiquitous intelligence, so that the interpretation power and applicability of the model in a target scene are limited. Traditional static social trust models, when faced with this complex scenario, expos