CN-122027344-A - Multidimensional dynamic trust evaluation method and system in heterogeneous embedded cluster environment

CN122027344ACN 122027344 ACN122027344 ACN 122027344ACN-122027344-A

Abstract

The invention discloses a multidimensional dynamic trust evaluation method and system in a heterogeneous embedded cluster environment, wherein the evaluation method comprises the steps of S1 obtaining historical interaction data, real-time resource data and third-party recommendation trust data of each node in the heterogeneous embedded cluster, S2 constructing a comprehensive behavior trust evaluation model for calculating a behavior trust value T1 of a target node, S3 constructing a capability trust evaluation model for calculating a capability trust value T2 of the target node, and S4 constructing a comprehensive decision model for outputting an initial comprehensive trust value T0 so as to provide basis for task unloading and scheduling of the heterogeneous embedded cluster. The invention has reasonable design, is beneficial to improving the use safety and the utilization rate of computing resources, and can provide basis for unloading heterogeneous embedded clusters.

Inventors

ZHANG ZHIWEI
Bi Maoze
ZHAO ZIHENG
LI GUANGXIA
WANG JIANDONG
WANG KAIZE
TIAN JING
TIAN YIHE
JI JIANZHENG
LI HAIYANG

Assignees

西安电子科技大学青岛计算技术研究院

Dates

Publication Date: 20260512
Application Date: 20260326

Claims (10)

1. A multidimensional dynamic trust evaluation method in a heterogeneous embedded cluster environment is characterized by comprising the following steps: The method comprises the steps of S1, acquiring historical interaction data, real-time resource data and third-party recommendation trust data of all nodes in a heterogeneous embedded cluster, wherein the historical interaction data at least comprise task completion states and task actual completion times, the real-time resource data at least comprise node residual energy and calculation loads, and the third-party recommendation trust data is trust evaluation of other nodes in the cluster on a target node; S2, constructing a comprehensive behavior trust evaluation model, calling historical interaction data of a target node and third-party recommendation trust data in S1, and calculating a behavior trust value T1 of the target node, wherein the comprehensive behavior trust evaluation model comprises a direct trust calculation module, a recommendation trust filtering module and a dynamic updating module, the direct trust calculation module constructs and outputs a target node direct trust value T11 based on the target node historical interaction data, the recommendation trust filtering module constructs and outputs a recommendation trust value T12 based on the third-party recommendation trust data, the direct trust value T11 and the recommendation trust value T12 are weighted and aggregated to obtain an initial behavior trust value T10, and the dynamic updating module is used for correcting the initial behavior trust value T10 to output the behavior trust value T1; s3, constructing a capability trust evaluation model, and calling real-time resource data in S1 independently of the comprehensive behavior trust evaluation model to calculate a capability trust value T2 of a target node, wherein the capability trust evaluation model comprises a multidimensional resource evaluation module and an objective weighting module; S4, constructing a comprehensive decision model, calling the comprehensive behavior trust evaluation model, calling the capability trust evaluation model, outputting an initial comprehensive trust value T01, acquiring a current task, constructing a feature vector of the current task, dynamically distributing weights according to the feature vector and aiming at a behavior trust value T1 and a capability trust value T2, outputting the initial comprehensive trust value T01 of a target node, calculating a final comprehensive trust value T0, finishing node trust evaluation, and providing a basis for task unloading and scheduling of heterogeneous embedded clusters.
2. The method for multidimensional dynamic trust assessment in heterogeneous embedded cluster environment of claim 1, wherein the comprehensive decision model further incorporates a dynamic security admission baseline Performing safety fusing, wherein the dynamic safety access baseline Regarding the number of interactions Is as follows: for trust maturation rate factors, controlling the rate at which admission criteria are tightened; for a lower initial latitude to accommodate new nodes; is a higher standard baseline to constrain the mature nodes; e is a natural constant (euler number); Introducing step penalty functions And the step penalty function Is designed as follows: Is that ; To blow the penalty factor, when a node triggers a security fuse, Preserving a value of trust assigned to the system; When the behavior trust value T1 calculated by the target node comprehensive behavior trust evaluation model is lower than the current dynamic security access baseline Triggering punishment, namely strongly setting the initial comprehensive trust value to 0 and taking the initial comprehensive trust value as a final comprehensive trust value T0, wherein when the behavior trust value T1 calculated by the target node comprehensive behavior trust evaluation model is not smaller than the current dynamic security access baseline And outputting the initial comprehensive trust value as a final comprehensive trust value T0 without triggering punishment.
3. The method for multidimensional dynamic trust evaluation in heterogeneous embedded cluster environment according to claim 2, wherein the direct trust calculation module outputs a direct trust value T11, comprising the steps of: S21, calling historical data of each node, and calculating timeliness score of single interaction of each node The age score Actual completion time with the single interaction Maximum tolerable delay Is related to the deviation of (2); s22, calling node history data to obtain the interaction success rate of each node; s23, calling average aging score of each node And multiplying the interaction success rate with the corresponding interaction success rate to obtain the effective service contribution degree of the target node, and outputting a direct trust value T11 by the direct trust calculation module based on the effective service contribution degree of the target node.
4. The method for multidimensional dynamic trust evaluation in heterogeneous embedded cluster environment according to claim 3, wherein the recommendation trust filter module outputs a recommendation trust value T12, comprising the steps of: S24, calling the third-party recommended trust data, and calculating cosine similarity of evaluation trend between each recommended node in the third-party recommended trust data and the target node; s25, setting a similarity threshold : For trust maturation rate factors, controlling the rate at which admission criteria are tightened; minimum value of similarity threshold; maximum value of similarity threshold; S26, removing outlier data, and enabling cosine similarity to be lower than a similarity threshold value The recommended node data of (1) is judged to be abnormal and is removed; And S27, carrying out weighted aggregation on the rest recommended trust data to obtain a recommended trust value T12.
5. The method for multidimensional dynamic trust assessment in heterogeneous embedded cluster environment according to claim 4, wherein the dynamic update module construction is based on node behavior fluctuation degree Asymmetric regulatory factor of (a) Function: e is a natural constant (euler number); t is time, representing the update time of the current behavior trust value, and keeping the real-time interaction time of the corresponding target node consistent with the time dimension of the node historical interaction sequence; alpha is a punishment adjustment coefficient and is a preset constant parameter, and the alpha is used for adjusting and controlling the increasing rate of the trust value when the node performance is superior to that of the history record; Beta is a punishment adjustment coefficient, is a preset constant parameter and is used for adjusting and controlling the descending rate of the trust value when the node is inferior to the history.
6. The method for multidimensional dynamic trust evaluation in heterogeneous embedded cluster environment of claim 5, wherein the degree of behavior fluctuation And (3) adopting any one of a difference absolute value method, a variance method of a history interaction sequence, a standard deviation or a behavior entropy method to obtain the target product.
7. The method for multidimensional dynamic trust evaluation under heterogeneous embedded cluster environment according to any one of claims 1-6, wherein the evaluation vector for evaluating physical capacity of the node is: Is a node Residual energy (revenue type); Is a node Computational load (consumption type); Is a node Communication bandwidth (revenue type).
8. The method for multi-dimensional dynamic trust evaluation under heterogeneous embedded cluster environment according to claim 7, wherein the weighting module assigns objective weights to components in the evaluation vector by using any one of a maximum dispersion method, an entropy weighting method and a CRITIC method.
9. The method for multidimensional dynamic trust evaluation in heterogeneous embedded cluster environment according to claim 7, wherein task feature vectors are constructed in the comprehensive decision model: The security sensitivity weight is the weight of the behavior trust value T1; The performance sensitivity weight is the weight of the capability trust value T2; And is also provided with =1; The comprehensive trust value T0=output by the comprehensive decision model *T1+ *T2。
10. A multidimensional dynamic trust evaluation system in a heterogeneous embedded cluster environment, the system being configured to implement the method of any one of claims 1-6, 8, 9, comprising: The data acquisition module is used for acquiring historical interaction data, real-time resource data and third-party recommendation trust data of each node in the heterogeneous embedded cluster; The behavior trust evaluation module is internally provided with a comprehensive behavior trust evaluation model and is used for calculating a behavior trust value T1 of the target node; The capability trust evaluation module is internally provided with a capability trust evaluation model, and is used for calling the real-time resource data in s1 and calculating a capability trust value T2 of the target node; And the comprehensive decision module is internally provided with a comprehensive decision model and is used for dynamically adjusting the aggregation weight and outputting a final comprehensive trust value T0 of the target node.

Description

Multidimensional dynamic trust evaluation method and system in heterogeneous embedded cluster environment Technical Field The invention relates to the technical field of computer computing resource allocation, in particular to a multidimensional dynamic trust evaluation method and system in a heterogeneous embedded cluster environment. Background With the vigorous development of the fields of edge computing, internet of things, autopilot and the like, a traditional centralized cloud computing mode is gradually extended to the edge of a network, and a heterogeneous embedded cluster formed by multiple types of computing units is gradually formed into a core computing architecture, wherein various embedded devices with different computing architectures, different computing power levels and storage capacities are integrated into the architecture. By virtue of the advantages of low delay, high bandwidth utilization rate and flexible deployment, the architecture is widely applied to scenes of processing complex calculation tasks such as automatic driving collaborative awareness, intelligent manufacturing real-time control, intelligent city security monitoring and the like. In addition, the architecture can optimize the configuration of computing resources according to the specific requirements of tasks, thereby remarkably improving the processing performance and the energy efficiency ratio of the system. Unlike cloud data centers, heterogeneous embedded clusters are limited by open environments and severe resource constraints (e.g., battery, computational power limitations), are difficult to run strong encryption protocols at full time, and are extremely vulnerable to black hole, doS, collusion, and the like. In addition, the inherent variability of the hardware units and the dynamic change characteristics of the resources also bring severe challenges to task scheduling and data transmission, and unreasonable scheduling strategies are extremely prone to performance degradation. To address the challenges described above, trust assessment is becoming a solution, relying primarily on statistical analysis and probabilistic prediction of node historic interaction records. The core calculates the direct trust value by counting the successful times and the failed times of task interaction. In the absence of direct interaction experience, such methods often introduce recommendation feedback from third party nodes, aggregating direct trust with recommendation trust by static linear weighting or simple averaging to build a global trust view. In addition, to reflect the timeliness of trust, conventional methods typically introduce a time decay factor that performs linear forgetting processing on historical trust values to reduce the impact of stale data on current evaluations. However, the existing trust evaluation mechanism faces the following three major core issues: The existing model is difficult to defend against joint attacks with strong concealment and changeable strategies. In a complex heterogeneous cluster environment, the evaluation mode is difficult to perceive service delay caused by malicious delay, and denial of service attack (DoS) cannot be effectively identified; Lack of resource awareness results in unjustification and resource waste. The existing scheme has the defect of 'one-cut', and cannot distinguish 'objective insufficient capacity' and 'subjective malicious behavior' caused by electric quantity exhaustion or calculation overload of nodes. The decision logic can cause that honest nodes with limited temporary resources are wrongly punished and removed, so that unnecessary reduction of available resources of the cluster is caused, and the utilization rate of the whole resources is seriously reduced; 3) The trust data dimension is single, the existing model generally only outputs a single trust scalar coupled with multiple factors, specific differences of nodes on the behavior security feature and the physical resource attribute are covered, the refined unloading block cannot be supported, and the precise scheduling requirement cannot be met. Disclosure of Invention The invention discloses a multidimensional dynamic trust evaluation method and a multidimensional dynamic trust evaluation system in a heterogeneous embedded cluster environment, which solve the technical problems that in the prior art, the security evaluation factor is single, the utilization rate of computing resources is low, and the computing power cannot be accurately scheduled. The technical scheme adopted is as follows: A multidimensional dynamic trust evaluation method under heterogeneous embedded cluster environment comprises the following steps: The method comprises the steps of S1, acquiring historical interaction data, real-time resource data and third-party recommendation trust data of all nodes in a heterogeneous embedded cluster, wherein the historical interaction data at least comprise task completion states and task actual