A multi-source trust-driven computing offloading and resource allocation method and system

Abstract

The application discloses a multi-source trust-driven computing offloading and resource allocation method and system, relates to the fields of cloud edge-end computing, network security and trust management, and comprises the following steps: a terminal user generates computing task information and sends the computing task information to a cloud service center; the cloud service center collects state information of each edge server; trust values of the edge servers are calculated from three dimensions of the cloud service center, the terminal user and the cooperative edge server respectively, so that three types of basic trust values, i.e., cloud-end trust, user-end trust and edge-end trust, are obtained; a comprehensive trust value is formed based on confidence weighting fusion; the trust value is dynamically updated; a trust perception net income maximization optimization problem is constructed; a resource allocation scheme is solved based on a Lagrange multiplier distribution method; and a computing offloading scheme is solved based on a trust perception greedy offloading method. The application realizes deep fusion of trust and offloading decision, and can effectively inhibit participation of dishonest edge nodes.

Description

Technical Field

[0001] This invention relates to the fields of cloud-edge computing, network security, and trust management, and in particular to a multi-source trust-driven method and system for computing offloading and resource allocation. Background Technology

[0002] With the deep integration of the Internet of Things (IoT) and artificial intelligence (AI), new applications with stringent requirements for latency and bandwidth, such as autonomous driving, industrial internet, and virtual reality, are constantly emerging. However, the massive amounts of data generated by these applications lead to information overload and network congestion in core networks. While cloud computing possesses ample computing resources, it cannot meet the real-time demands of data interaction and processing due to network latency and bandwidth constraints. To address this challenge, edge computing has emerged. Edge computing can push computing, storage, and network resources to the network edge, thereby providing services closer to end users, effectively shortening service response time and reducing network load. However, edge devices are often lightweight, and limited resources are an inherent limitation.

[0003] Cloud-Edge-End Computing (CEEC) systems integrate cloud computing and edge computing, providing ample computing resources while meeting real-time requirements and alleviating bandwidth pressure. In CEEC, end users (EUs) with limited computing power migrate some or all of their computationally intensive tasks to more resource-rich edge servers (ES) or cloud service centers (CSCs) for processing. This task offloading helps minimize latency and improve system service quality in resource-constrained environments.

[0004] In CEEC's computation offloading and resource allocation, customizing offloading decisions based on system state information is crucial. Existing research has achieved effective offloading strategies by optimizing metrics such as resources and latency using mathematical optimization algorithms or machine learning. However, due to CEEC's distributed and open nature, the offloading process faces node security challenges: selfish / malicious edge nodes may refuse to cooperate, provide low-quality services, and provide false information. Dishonest behavior by edge nodes severely impacts CEEC's security, reliability, and service efficiency.

[0005] Existing technical solution 1: Wang et al. (IEEE Transactions on Mobile Computing, 2025) proposed a trust-based computation offloading framework (TCO) for mobile cloud-edge computing networks. This method employs a two-layer trust evaluation, solving for the offloading strategy through stochastic adaptive search and gradient descent. Although TCO introduces a trust factor, its trust sources only include the cloud and user layers, and it lacks an effective trust update mechanism, resulting in slow trust convergence and insufficient responsiveness to dynamic attacks. Furthermore, trust is only used for gain calculation and resource discounting, and is not deeply integrated into the optimization objective.

[0006] Existing technical solution 2: Guo et al. (IEEE Transactions on Cognitive Communications and Networking, 2024) proposed a Trusted and Efficient Task Offloading (FCTE) method for vehicular edge computing networks. This method uses a fuzzy comprehensive evaluation method for trust calculation, but its trust updates are mainly based on a fixed period, resulting in a lag in response to dynamic attacks and an inability to quickly identify collusion attacks.

[0007] The aforementioned technical solutions, as representative of existing solutions, mostly suffer from limitations such as a single source of trust, shallow coupling between trust and offloading decisions, and a lack of effective trust update mechanisms. Therefore, there is an urgent need for a multi-source trust-driven computational offloading and resource allocation method and system that can address these technical problems. Summary of the Invention

[0008] This solution addresses the problems and needs raised above by proposing a multi-source trust-driven computation offloading and resource allocation method and system. It achieves the above-mentioned technical objectives and brings about several other technical benefits by adopting the following technical features.

[0009] One objective of this invention is to propose a multi-source trust-driven computation offloading and resource allocation method, comprising the following steps: S10: The end user generates computing task information and sends it to the cloud service center. The computing task information includes the amount of input data, the required computing capacity, and the maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge server. S20: Calculate the trust value of the edge server from three dimensions: cloud service center, end user, and collaborative edge server, to obtain the cloud trust value. User trust and edge trust Three basic trust values; S30: Based on the confidence-weighted fusion method, the three types of basic trust values ​​are fused into a comprehensive trust value. ; S40: Employs an update mechanism based on the Ebbinghaus forgetting curve to adjust the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; S50: Construct a trust-perceived net benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation subproblems and computation unloading subproblems by incorporating the delay model and cost function; the resource allocation subproblem is solved based on the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task; S60: Solve the computational unloading subproblem based on a trust-aware greedy unloading method, processing user-requested tasks sequentially according to their urgency, and using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

[0010] In addition, the multi-source trust-driven computation offloading and resource allocation method and system according to the present invention may also have the following technical features: In one example of the present invention, in step S20, cloud trust... The calculation process is as follows: S21: Calculate the task timeout rate of the edge server. That is, the proportion of timed-out tasks to the total number of tasks; S22: Calculate service consistency by comparing task completion times in edge server and end-user records. Its expression is:

[0011] In the formula, N is the length of the service record; For the i-th service time record of the end user; This is the i-th service time record for the edge server; S23: Obtain a cloud trust score by weighting task timeout rate and service consistency. Its expression is:

[0012] In the formula, (▪) is the natural exponential function.

[0013] In one example of the present invention, in step S20, edge trust is calculated. The calculation process is as follows: S201: When the primary edge server assigns a task to a collaborating edge server for collaborative processing, both parties record the collaborative response time. ; Calculate response timeliness Its expression is:

[0014] In the formula, This is the time decay parameter; Among them, the time decay parameter is used to control the impact of collaborative response time on response timeliness, so that the longer the response time, the smaller the increment. S202: Calculate Collaboration Familiarity This refers to the number of collaborations provided to a specific edge server. The percentage of total collaborations is expressed as follows:

[0015] In the formula, This refers to the number of collaborations provided to other edge servers; The number of edge servers that have collaborated; i is the iteration variable; S203: Edge Trust Score The weighted fusion of response timeliness and collaboration familiarity is expressed as follows:

[0016] In the formula, Let represent the collaboration quality factor of the i-th collaborative edge server relative to edge server m.

[0017] In one example of the present invention, in step S20, the user trusts... The calculation method is as follows: End users form trust ratings based on their direct interaction records with the edge server; after each interaction, end users record the actual task completion time. Compared with the expected completion time Deviation; defining user-side trust Its expression is as follows:

[0018] In the formula, The actual time to complete the task; This refers to the expected completion time of the task.

[0019] In one example of the present invention, step S30 specifically includes the following steps: S31: Treat the cloud service center, end users, and collaborative edge servers as different sources of trust; for each source of trust, the confidence level... The product of three factors:

[0020] In the formula, For data volume factor, As a stability factor, Freshness factor; This represents the cumulative number of tasks corresponding to the trusted source. The variance of the trust value corresponding to the trust source; This refers to the latest update time corresponding to the trusted source; Among them, data volume factor Based on the number of interactions, the specific expression is:

[0021] In the formula, To control parameters; regarding cloud trust, This refers to the cumulative number of tasks; regarding user trust, The number of end users participating in the evaluation; for edge trust, The number of edge servers that have collaborated; Among them, stability factor The variance calculation based on trust values ​​is specifically expressed as follows:

[0022] In the formula, For control parameters; The variance of the trust value; Among them, freshness factor Calculated based on the time of the most recent update;

[0023] In the formula, The current moment; This is the latest update time from the trusted source; This is the time decay parameter; S32: Overall Trust Value Defined as a weighted average of the confidence values, with the weights being the corresponding confidence levels:

[0024] In the formula, The confidence level of a cloud trust source; Assess trust in the cloud; The confidence level of edge trust sources; Trust scoring at the edge; The number of edge servers that have collaborated; The confidence level of the end-user's trusted source; The number of end users served; Assess user trust rating; This represents the set of edge servers that have collaboration records with edge server m; Represents a set Any collaborative edge server in the network; This represents the set of end users served by edge server m; Represents a set The nth terminal user in the process.

[0025] In one example of the present invention, in step S40, an update mechanism based on the Ebbinghaus forgetting curve is used to update the overall trust value. Dynamic updates involve the following steps: S41: Suppose that the edge server has gone through J evaluation cycles since the initial time, and obtains a comprehensive trust value at the end of each cycle. And record the time. For the first Trust value over a period of time, and its forgetting factor for:

[0026] In the formula, The time interval between two updates; and These are the fitting parameters; S42: The current overall trust value is a weighted average of historical trust values.

[0027] In the formula, For the first Period and the The time interval of each cycle.

[0028] In one example of the present invention, step S50 specifically includes the following steps: S51: Set the task delay equal to the maximum tolerable delay. Find the minimum computing resources required while ensuring the latency requirement. Its expression is:

[0029] In the formula, The computing capacity required to complete the task The overall trust value for edge servers; Size of the input data; The uplink transmission rate for users to upload tasks to the edge server; S52: Introduction of Lagrange multipliers The constrained optimization problem is transformed into an unconstrained Lagrangian function extremum problem, specifically expressed as:

[0030] In the formula, Cost of computing resources per unit for edge servers; The amount of computing resources committed to the edge server for the task; This is the revenue conversion factor; The computing capacity required to complete the task; As a comprehensive trust value; Idle computing resources for edge servers; This represents the nth end user or its corresponding computing task that is offloaded to edge server m; Represents the Lagrange multiplier corresponding to the resource constraints of edge server m; S53: For each task, the candidate resource allocation value is obtained by ensuring that the partial derivatives are zero and that the complementary relaxation condition is satisfied. The expression is as follows:

[0031] In the formula, Lagrange multipliers that satisfy the Karush-Kuhn-Tucker conditions; S54: Searching for the optimal Lagrange multiplier using the binary search method The following condition is true, and the specific expression is:

[0032] In the formula, To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met.

[0033] In one example of the present invention, step S60 specifically includes the following steps: S61: Generate all feasible ES solutions for each task, following trust conditions, subband conditions, transmission feasibility and resource feasibility constraints; S62: Sort tasks in descending order of urgency, where the urgency is determined by the task's computation density 2. The higher the urgency value, the higher the priority of the task. S63: Calculate the greedy value for each feasible solution to the current task. :

[0034] In the formula, Calculate the unit resource cost for the cloud service center; To determine the minimum computing resources required to offload tasks to the cloud service center while ensuring latency requirements are met; To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met; The remaining resources of the edge server at the current moment; Initial idle resources for edge servers; S64: Select the option with the largest greedy value as the unloading option for the current task; update the resource status and continue processing the next task.

[0035] Another objective of this invention is to propose a multi-source trust-driven computation offloading and resource allocation system, comprising: The task information collection module is configured to generate computing task information for end users and send it to the cloud service center. This computing task information includes the amount of input data, required computing capacity, and maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge servers. The multi-source trust assessment module is configured to calculate the trust value of the edge servers from three dimensions: the cloud service center, the end user, and the collaborating edge servers, respectively, to obtain the cloud-based trust value. User trust and edge trust Three basic trust values; The comprehensive trust fusion module is configured to fuse the three types of basic trust values ​​into a comprehensive trust value based on a confidence-weighted fusion method. ; The trust update module is configured to use an update mechanism based on the Ebbinghaus forgetting curve to update the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; The resource allocation module is configured to construct the Trust-Aware Net Benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation subproblems and computation unloading subproblems by incorporating the delay model and cost function; the resource allocation subproblem is solved based on the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task; The computational unloading module is configured to solve the computational unloading subproblem based on a trust-aware greedy unloading method. It processes user-requested tasks sequentially according to their urgency, using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

[0036] In one example of the present invention, the multi-source trust assessment module includes: Cloud-based trusted computing unit, configured for task timeout rate based on edge server Service Consistency Calculate cloud trust scores ; The edge-side trust computing unit is configured to provide timely responses based on collaborative edge servers. and familiarity with collaboration Calculate edge trust score ; The user-side trust calculation unit is configured to calculate the actual task completion time based on the interaction records between the end user and the edge server. Compared with the expected completion time Deviation calculation of user trust score .

[0037] Compared with the prior art, the present invention has the following beneficial effects: 1. The multi-source trust model proposed in this invention quantifies the trustworthiness of edge servers from three dimensions: cloud service center, end user, and collaborative edge server. It also combines the forgetting curve to realize the dynamic update of trust value, thus solving the problems of single trust source and lack of trust update mechanism in the prior art.

[0038] 2. This invention deeply integrates trust factors into the entire process of computation unloading and resource allocation, constructs a trust-perceived net benefit (TANB) maximization optimization objective, and effectively suppresses the participation of dishonest nodes through trust penalties.

[0039] 3. The joint optimization scheme of the Lagrange multiplier allocation method and the trust-aware greedy unloading method proposed in this invention achieves accurate solution of resource allocation and efficient selection of unloading scheme while ensuring computational efficiency.

[0040] 4. This invention outperforms existing methods in terms of trust perception net benefit and trust convergence performance, and exhibits higher accuracy and faster convergence in forgery attack and collusion attack scenarios.

[0041] The preferred embodiments of the invention will be described in more detail below with reference to the accompanying drawings, so as to facilitate an understanding of the features and advantages of the invention. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. The drawings are merely illustrative of some embodiments of the present invention and are not intended to limit the scope of the present invention to all embodiments.

[0043] Figure 1 A flowchart illustrating the multi-source trust-driven computation offloading and resource allocation method provided in this embodiment of the invention; Figure 2 This is a comparison chart of net trust perception benefits under different numbers of terminal users provided in an embodiment of the present invention; Figure 3 This is a comparison chart of trust-aware net benefits under different numbers of edge servers provided in an embodiment of the present invention; Figure 4 This is a comparison chart of trust-aware net benefits under different idle computing resources provided in an embodiment of the present invention; Figure 5 This is a trust convergence curve under a forgery attack provided in an embodiment of the present invention; Figure 6 This is a trust convergence curve diagram for the conversion of a dishonest node into an honest node, provided in an embodiment of the present invention. Figure 7 A comparison chart of trust convergence curves under collusion attacks provided in embodiments of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The same reference numerals in the drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0045] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, “an” or “a” and similar terms do not necessarily indicate a quantity limitation. Terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships; these relative positional relationships may change accordingly when the absolute position of the described object changes.

[0046] According to a first aspect of the present invention, a multi-source trust-driven computation offloading and resource allocation method is provided, such as... Figure 1 As shown, it includes the following steps: S10: The end user generates computing task information and sends it to the cloud service center. The computing task information includes the amount of input data, the required computing capacity, and the maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge server. S20: Calculate the trust value of the edge server from three dimensions: cloud service center, end user, and collaborative edge server, to obtain the cloud trust value. User trust and edge trust Three basic trust values; S30: Based on the confidence-weighted fusion method, the three types of basic trust values ​​are fused into a comprehensive trust value. ; S40: Employs an update mechanism based on the Ebbinghaus forgetting curve to adjust the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; S50: Construct a trust-perceived net benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation subproblems and computation unloading subproblems by incorporating the delay model and cost function; the resource allocation subproblem is solved based on the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task; S60: Solve the computational unloading subproblem based on a trust-aware greedy unloading method, processing user-requested tasks sequentially according to their urgency, and using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

[0047] The proposed multi-source trust model quantifies the trustworthiness of edge servers from three dimensions: cloud service center, end user, and collaborative edge server. It also incorporates a forgetting curve to dynamically update trust values, thus solving the problems of a single trust source and a lack of trust update mechanism in existing technologies.

[0048] This method deeply integrates trust factors into the entire process of computation unloading and resource allocation, constructs a trust-aware net benefit (TANB) maximization optimization objective, and effectively suppresses the participation of dishonest nodes through trust penalties.

[0049] The proposed method combines the Lagrange multiplier allocation method and the trust-aware greedy unloading method to achieve accurate resource allocation and efficient selection of unloading schemes while ensuring computational efficiency.

[0050] This method outperforms existing methods in terms of trust-aware net benefit and trust convergence performance, and exhibits higher accuracy and faster convergence in forgery attack and collusion attack scenarios.

[0051] In one example of the present invention, in step S20, cloud trust... The calculation process is as follows: S21: Calculate the task timeout rate of the edge server. That is, the proportion of timed-out tasks to the total number of tasks; S22: Calculate service consistency by comparing task completion times in edge server and end-user records. Its expression is:

[0052] In the formula, N is the length of the service record; For the i-th service time record of the end user; This is the i-th service time record for the edge server; S23: Obtain a cloud trust score by weighting task timeout rate and service consistency. Its expression is:

[0053] In the formula, (▪) is the natural exponential function.

[0054] In one example of the present invention, in step S20, edge trust is calculated. The calculation process is as follows: S201: When the primary edge server assigns a task to a collaborating edge server for collaborative processing, both parties record the collaborative response time. ; Calculate response timeliness Its expression is:

[0055] In the formula, This is the time decay parameter; Among them, the time decay parameter is used to control the impact of collaborative response time on response timeliness, so that the longer the response time, the smaller the increment. S202: Calculate Collaboration Familiarity This refers to the number of collaborations provided to a specific edge server. The percentage of total collaborations is expressed as follows:

[0056] In the formula, This refers to the number of collaborations provided to other edge servers; The number of edge servers that have collaborated; i is the iteration variable; S203: Edge Trust Score The weighted fusion of response timeliness and collaboration familiarity is expressed as follows:

[0057] In the formula, This represents the collaboration quality factor of the i-th collaborative edge server relative to edge server m, used to characterize the reliability of collaboration of this collaborative edge server in the historical collaboration process.

[0058] In one example of the present invention, in step S20, the user trusts... The calculation method is as follows: End users form trust ratings based on their direct interaction records with the edge server; after each interaction, end users record the actual task completion time. Compared with the expected completion time Deviation; defining user-side trust Its expression is as follows:

[0059] In the formula, The actual time to complete the task; This refers to the expected completion time of the task.

[0060] In one example of the present invention, step S30 specifically includes the following steps: S31: Treat the cloud service center, end users, and collaborative edge servers as different sources of trust; for each source of trust, the confidence level... The product of three factors:

[0061] In the formula, For data volume factor, As a stability factor, Freshness factor; This represents the cumulative number of tasks corresponding to the trusted source. The variance of the trust value corresponding to the trust source; This refers to the latest update time corresponding to the trusted source; Among them, data volume factor Based on the number of interactions, the specific expression is:

[0062] In the formula, To control parameters; regarding cloud trust, This refers to the cumulative number of tasks; regarding user trust, The number of end users participating in the evaluation; for edge trust, The number of edge servers that have collaborated; Among them, stability factor The variance calculation based on trust values ​​is specifically expressed as follows:

[0063] In the formula, For control parameters; The variance of the trust value; Among them, freshness factor Calculated based on the time of the most recent update;

[0064] In the formula, The current moment; This is the latest update time from the trusted source; This is the time decay parameter; S32: Overall Trust Value Defined as a weighted average of the confidence values, with the weights being the corresponding confidence levels:

[0065] In the formula, The confidence level of a cloud trust source; Assess trust in the cloud; The confidence level of edge trust sources; Trust scoring at the edge; The number of edge servers that have collaborated; The confidence level of the end-user's trusted source; The number of end users served; Assess user trust rating; This represents the set of edge servers that have collaboration records with edge server m; Represents a set Any collaborative edge server in the network; This represents the set of end users served by edge server m; Represents a set The nth terminal user in the process.

[0066] In one example of the present invention, in step S40, an update mechanism based on the Ebbinghaus forgetting curve is used to update the overall trust value. Dynamic updates involve the following steps: S41: Suppose that the edge server has gone through J evaluation cycles since the initial time, and obtains a comprehensive trust value at the end of each cycle. And record the time. For the first Trust value over a period of time, and its forgetting factor for:

[0067] In the formula, The time interval between two updates; and These are the fitting parameters; S42: The current overall trust value is a weighted average of historical trust values.

[0068] In the formula, For the first Period and the The time interval of each cycle.

[0069] In one example of the present invention, step S50 specifically includes the following steps: S51: Set the task delay equal to the maximum tolerable delay. Find the minimum computing resources required while ensuring the latency requirement. Its expression is:

[0070] In the formula, The computing capacity required to complete the task The overall trust value for edge servers; Size of the input data; The uplink transmission rate for users to upload tasks to the edge server; S52: Introduction of Lagrange multipliers The constrained optimization problem is transformed into an unconstrained Lagrangian function extremum problem, specifically expressed as:

[0071] In the formula, Cost of computing resources per unit for edge servers; The amount of computing resources committed to the edge server for the task; This is the revenue conversion factor; The computing capacity required to complete the task; As a comprehensive trust value; Idle computing resources for edge servers; This represents the nth end user or its corresponding computing task that is offloaded to edge server m; Represents the Lagrange multiplier corresponding to the resource constraints of edge server m; S53: For each task, the candidate resource allocation value is obtained by ensuring that the partial derivatives are zero and that the complementary relaxation condition is satisfied. The expression is as follows:

[0072] In the formula, Lagrange multipliers that satisfy the Karush-Kuhn-Tucker conditions; S54: Searching for the optimal Lagrange multiplier using the binary search method The following condition is true, and the specific expression is:

[0073] In the formula, To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met.

[0074] In one example of the present invention, step S60 specifically includes the following steps: S61: Generate all feasible ES solutions for each task, following trust conditions, subband conditions, transmission feasibility and resource feasibility constraints; S62: Sort tasks in descending order of urgency, where urgency is determined by the task's computational density. The required computing capacity / maximum tolerable latency determines the urgency of the task; a higher urgency value indicates that the task should be processed first. S63: Calculate the greedy value for each feasible solution to the current task. :

[0075] In the formula, Calculate the unit resource cost for the cloud service center; To determine the minimum computing resources required to offload tasks to the cloud service center while ensuring latency requirements are met; To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met; The remaining resources of the edge server at the current moment; Initial idle resources for edge servers; S64: Select the option with the largest greedy value as the unloading option for the current task; update the resource status and continue processing the next task.

[0076] According to a second aspect of the present invention, a multi-source trust-driven computation offloading and resource allocation system includes: The task information collection module is configured to generate computing task information for end users and send it to the cloud service center. This computing task information includes the amount of input data, required computing capacity, and maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge servers. The multi-source trust assessment module is configured to calculate the trust value of the edge servers from three dimensions: the cloud service center, the end user, and the collaborating edge servers, respectively, to obtain the cloud-based trust value. User trust and edge trust Three basic trust values; The comprehensive trust fusion module is configured to fuse the three types of basic trust values ​​into a comprehensive trust value based on a confidence-weighted fusion method. ; The trust update module is configured to use an update mechanism based on the Ebbinghaus forgetting curve to update the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; The resource allocation module is configured to construct the Trust-Aware Net Benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation subproblems and computation unloading subproblems by incorporating the delay model and cost function; the resource allocation subproblem is solved based on the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task; The computational unloading module is configured to solve the computational unloading subproblem based on a trust-aware greedy unloading method. It processes user-requested tasks sequentially according to their urgency, using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

[0077] The multi-source trust model proposed by this system quantifies the trustworthiness of edge servers from three dimensions: cloud service center, end user, and collaborative edge server. It also combines the forgetting curve to realize the dynamic update of trust value, which solves the problems of single trust source and lack of trust update mechanism in the existing technology.

[0078] This system deeply integrates trust factors into the entire process of computation offloading and resource allocation, constructs a trust-aware net benefit (TANB) maximization optimization objective, and effectively suppresses the participation of dishonest nodes through trust penalties.

[0079] The proposed joint optimization scheme of the Lagrange multiplier allocation method and the trust-aware greedy unloading method achieves accurate solution of resource allocation and efficient selection of unloading scheme while ensuring computational efficiency.

[0080] This system outperforms existing methods in terms of trust-aware net benefit and trust convergence performance, and exhibits higher accuracy and faster convergence in forgery attack and collusion attack scenarios.

[0081] In one example of the present invention, the multi-source trust assessment module includes: Cloud-based trusted computing unit, configured for task timeout rate based on edge server Service Consistency Calculate cloud trust scores ; The edge-side trust computing unit is configured to provide timely responses based on collaborative edge servers. and familiarity with collaboration Calculate edge trust score ; The user-side trust calculation unit is configured to calculate the actual task completion time based on the interaction records between the end user and the edge server. Compared with the expected completion time Deviation calculation of user trust score .

[0082] It should be noted that the multi-source trust-driven computation offloading and resource allocation system of the present invention can also perform any of the processing described in the previously described multi-source trust-driven computation offloading and resource allocation method, and the specific details are not repeated here.

[0083] Specific examples: System Model: The cloud-edge-device computing (CEEC) system considered in this invention includes a cloud service center (CSC), an edge service layer, and an end-user (EU) layer. The CSC is responsible for scheduling and managing the computing and communication resources of the edge service (ES), and formulates a reasonable computation offloading and resource allocation scheme upon receiving service requests from the EUs. Each ES provides services to multiple EUs, and ESs can communicate with each other to complete collaborative tasks. Each EU has limited computing resources; when its local resources are insufficient, it sends a request to the CSC to offload computing tasks.

[0084] In this embodiment, the EU set and ES collection Together, they describe the unloading object and candidate compute nodes. After the end user submits a task to the CSC, the task information consists of the input data volume. Required computing capacity and maximum tolerable delay Characterization; CSC further acquires idle computing resources from ES and unit cost Cloud computing unit cost satisfy Therefore, edge processing should be prioritized when satisfying trust constraints. The main Elasticsearch (ES) can delegate some tasks to collaborating ESs, and the task splitting and collaborative execution relationship can be characterized by collaboration-related variables.

[0085] In this system, it is assumed that CSC and EU are completely trustworthy, and only the trustworthiness of ES is considered. Dishonest ES (DES) typically launch forgery attacks, falsely reporting their own idle resources and service records. Furthermore, the distributed collaborative system architecture is prone to attracting collusion attacks.

[0086] Optimization problem of perceived net benefit of trust: When making an uninstallation decision, the CSC also determines the user association. Subband association and resource allocation .in, This indicates that the task is handled by the CSC. This indicates that the task is unloaded to Elasticsearch; This represents the number of subbands available in a single ES, where subband occupancy satisfies... and The mutually exclusive relationship defined by the agreement; This represents the amount of computational resources that Elasticsearch (ES) has committed to the task. It also represents the set of tasks unloaded to the CSC. Task sets to be uninstalled to Elasticsearch and collaborative ES collections Together, they define the decision-making scope for subsequent TANB optimization.

[0087] To ensure that the optimization objective simultaneously reflects benefits, costs, and trust risks, this invention constructs the TANB maximization problem. This objective deducts cloud computing costs and edge resource costs adjusted for trust values ​​from the benefits of early task completion, thus penalizing low-trust Elasticsearch Entities (ES) during optimization.

[0088]

[0089]

[0090] In the optimization problem described above, constraint C1 states that the computational resources allocated to each task are non-negative. C2 and C3 state that the number of users served simultaneously by each ES cannot exceed the total number of its available subbands, and different EUs on the same ES must use different subbands. C4 states that the total computational resources allocated by an ES to all tasks must not exceed its idle computational resources. C5 is a latency constraint, representing the actual completion time of each task. It must not exceed its maximum tolerable delay. C6 and C7 ensure that the trust value of each ES participating in each stage does not fall below the threshold. To ensure basic reliability.

[0091] TANB consists of three parts: satisfaction gains, cloud computing costs, and edge resource costs. Satisfaction gains are related to the task's early completion time; cloud costs reflect the local processing costs of the CSC; and edge costs are determined by the overall trust value. Perform risk correction to ensure that even with low-reliability ES, the nominal resource price remains the same. Lower values ​​will also result in higher equivalent costs due to insufficient ability to fulfill obligations.

[0092] Delay model: This embodiment decomposes the task completion time into transmission delay and computation delay, and introduces trust correction into the computation delay. The transmission rate is determined by link gain, interference, and noise; the lower the trust value, the fewer effective computing resources that ES can reliably provide, and the longer the task completion time.

[0093] When a task is executed by CSC, the transmission latency to CSC and the cloud computing latency are added together to obtain the task completion time.

[0094]

[0095] When a task is executed by Elasticsearch (ES), the transmission latency to ES is added to the edge computing latency corrected by the trust value to obtain the task completion time.

[0096]

[0097] The upload rate to CSC is denoted as The upload rate to Elasticsearch is denoted as . Both are determined by the channel conditions and interference level of the corresponding link, respectively.

[0098] When a task is processed collaboratively by multiple Elasticsearch (ES) instances, the primary ES instance... Retain a portion of the computational load and allocate the rest to the cooperative ES. The computational load borne by the main Elasticsearch engine itself. , amount of data to retain The amount of computation undertaken by collaborative ES and the amount of data distributed The collaborative execution process is jointly determined; the main ES can wait for the collaborative results while performing its own calculations.

[0099]

[0100] Multi-source trust model: The multi-source trust model is used to simultaneously measure the reliability of ES from the management side, the user side, and the collaboration side, and integrates multiple trust sources into a comprehensive trust value according to their confidence level.

[0101] (a) Trust in the cloud Cloud trust is maintained by CSC based on ES's historical task execution and service records, and is used to reflect the overall reliability of ES in long-term system operation.

[0102] For service anomalies caused by false resource reporting, CSC uses ES... Task timeout rate Evaluate its reliability; the timeout rate is determined by the number of timeout tasks and the total number of tasks. The more frequent the timeouts, the lower the cloud score.

[0103] In cases where service records have been tampered with, CSC will record the number of users served by ES. ES side completion time record EU side completion time record and the Corresponding user records The comparison is performed, and service consistency is calculated based on the recorded deviations.

[0104]

[0105] when Service consistency at the time of establishment The maximum value is achieved; as the deviation between the records on both sides increases, service consistency decreases.

[0106] After considering the overall task timeout rate and service consistency, a cloud trust score is obtained from the CSC side: :

[0107] (ii) Edge Trust Edge trust is built upon the collaboration records between Elasticsearch Entities (ES). The current ES, other ES collaborating with it, and the ES being evaluated together constitute the edge-side evaluation relationship. After the main ES and collaborating ES complete their task interactions, they update each other's trustworthiness based on response latency and collaboration frequency.

[0108] Response timeliness is determined by collaborative response time. The resulting response timeliness score is obtained through mapping. The higher the value, the faster the collaborative ES will react from receiving a request to starting execution.

[0109]

[0110] The time decay parameter maps response latency to trust increments; the longer the response latency, the smaller the increment. Collaboration familiarity measures the proportion of participation in historical collaborations for a given Elasticsearch instance, which can be determined by the number of collaborations. Total number of collaborations and corresponding familiarity The results are calculated and weighted to reflect the reference value of the evaluation source.

[0111]

[0112] Using a logarithmic approach to handle collaboration frequency can reduce the unilateral influence of extremely high-frequency collaboration nodes on the results, making the weights of different collaboration sources more balanced.

[0113] By combining responsiveness and familiarity with collaboration, an edge trust score is obtained:

[0114] (iii) User-side trust User trust is built by EU based on direct interactive experience. EU compares actual task completion times. Compared with the expected completion time The greater the deviation between the two, the lower the user rating.

[0115] The user trust score for a single interaction is determined by the deviation between the actual completion time and the expected completion time.

[0116]

[0117] User Trust Rating It is between 0 and 1; it is satisfied when the actual completion time is close to the expected time. The score is relatively high, but when the timeout is severe, the score approaches 0.

[0118] (iv) Comprehensive Trust Overall trust is used to represent the overall reliability of ES when performing tasks. This invention uses CSC, EU, and cooperative ES as different trust sources, corresponding to a basic trust value. , and And combined with the confidence levels of each trust source and overall trust value Complete weighted fusion.

[0119]

[0120] In the formula, For data volume factor, As a stability factor, Freshness factor; This represents the cumulative number of tasks corresponding to the trusted source. The variance of the trust value corresponding to the trust source; This refers to the latest update time corresponding to the trusted source; Data volume factor The stability factor is determined based on the number of available samples for the corresponding trust source. The freshness factor is determined based on the fluctuation of trust value. Determined based on the most recent update time; cumulative task records correspond to the cloud, user terminal, and edge terminal respectively. Other evaluation sample size .

[0121]

[0122] Among them, control parameters , and These are used to adjust the impact of data volume, stability, and freshness on confidence levels, respectively.

[0123] The overall ES trust score is a weighted average based on the confidence level of each trust source:

[0124] When the confidence level of all trust sources is zero, a preset default trust value can be used; continuous confidence weights can avoid abrupt changes caused by hard thresholds.

[0125] (v) Trust Renewal Since the representativeness of early behavior to current credibility gradually decreases, this invention uses a forgetting curve to weaken the influence of old trust records, so that recent observations have a higher weight in the overall trust update.

[0126]

[0127] Among them, time interval and fitting parameters , Used to control the decay rate; parameters and It can be preset or adjusted according to the system's reliance on historical records.

[0128] Let ES exist The historical comprehensive trust value is formed within each assessment cycle; The historical trust value for each period is recorded as follows: The time of its occurrence is The trust value for the current period is denoted as... The current time is The time interval between the two is Based on this, the forgetting factor is calculated. The current overall trust is obtained by weighting historical trust.

[0129]

[0130] The above update method assigns lower weight to older records; initial trust value and forgetting factor This collectively ensures that recent behavior has a greater impact on the final trust score.

[0131] Optimize the problem-solving approach: (a) Problem Breakdown TANB is a hybrid optimization problem involving discrete unloading decisions and continuous resource allocation. To reduce the difficulty of solving this problem, this invention first fixes the unloading relationship to solve for resource allocation, and then determines the unloading scheme under the condition of resource feasibility.

[0132] For tasks offloaded to Elasticsearch, the task completion time can be allowed to reach the maximum tolerable latency boundary first, thereby obtaining the minimum computing resources required to meet the latency requirements.

[0133]

[0134] As long as the actual allocated resources are not lower than the minimum resources and the total allocated amount does not exceed the ES idle resources, the delay constraint can be transformed into the following resource constraint form.

[0135]

[0136] Once the unloading scheme is fixed, the resource allocation on each ES can be decomposed into mutually independent subproblems. The optimal value of each subproblem corresponds to the equivalent cost after minimizing the resource purchase cost and the task computation delay loss.

[0137]

[0138]

[0139] Since the transmission time is fixed under the fixed offloading scheme, maximizing TANB can be equivalently transformed into minimizing the aforementioned resource-related costs; the trust value affects both cost adjustment and minimum resource requirements.

[0140] Once the lower bound of the resources is determined, the remaining problem can be represented as a computational unloading subproblem:

[0141]

[0142] In the offloading decision, low-trust ES is affected by overall trust. Lower costs result in a cost penalty, and are due to minimum resource requirements. The rise is suppressed; candidate solutions also need to meet individual trust thresholds and comprehensive trust thresholds.

[0143] (ii) Lagrange Multiplier Allocation Method (LMAM) For the resource allocation subproblem, this invention uses the Lagrange multiplier allocation method to solve it.

[0144] Given the unloading scheme, the main ES and sub-bands corresponding to each task are determined. Resource allocation can be regarded as a continuous optimization problem within the main ES and written in the following form.

[0145]

[0146]

[0147] This resource allocation problem is a convex problem, constrained by a linear resource ceiling. The objective function consists of the resource payment cost and computational delay loss. Increasing resources will reduce delay loss but increase payment cost; therefore, the optimal point reflects the balance between marginal revenue and marginal cost.

[0148] To address the total resource constraints of ES, Lagrange multipliers are introduced, transforming the resource constraints coupled between multiple tasks into a problem of solving individual multipliers.

[0149]

[0150] Lagrange multipliers satisfying KKT conditions Corresponding to the globally optimal resource allocation, for each task The candidate resource solution can be obtained by taking the partial derivative.

[0151]

[0152] Simultaneously, the complementary relaxation condition must be satisfied:

[0153] The candidate resource allocation expression can be obtained from the partial derivative relation:

[0154] when When taking a given value, if the candidate resource solution All resources provided are at least as low as the lower bound. And the total does not exceed the ES free resources If the solution is correct, then it can be used directly; otherwise, increase the value. Binary search is used to make the total number of resources equal to... and obtain the conditions that are met. .

[0155]

[0156] The overall trust value is used in the candidate resource calculation. A lower trust value corresponds to a higher equivalent resource requirement or cost, thus limiting the resource competition.

[0157] Tasks processed by CSC do not participate in edge-side resource contention, and their optimal resources can be directly calculated:

[0158] (III) Trust Perception Greedy Unloading Method (TGOM) The computational offloading subproblem is used to select the most profitable combination among feasible Elasticsearch (ES), subband, and cloud processing solutions. Since the candidate combinations grow rapidly with the number of tasks and ES, this invention employs a trust-aware greedy offloading method to process tasks one by one and dynamically update resource status.

[0159] First, construct a set of candidate solutions for each task. Each candidate solution contains a main ES. Sub-band and minimum resource requirements It also meets the requirements of trust threshold, subband availability, transmission feasibility, and resource lower bound.

[0160] Subsequently, based on the calculated density Determine the urgency of each task and generate a sequence of tasks to be processed in descending order of urgency. .

[0161] While processing the current task, the system reads the remaining resources from each Elasticsearch instance. Sub-band occupancy status , select currently executable solutions from the initial candidate set; If edge processing solution is selected The minimum resource requirement It should not exceed the current remaining resources And the selected sub-band satisfies the condition of not being occupied. .

[0162] If the CSC processing solution is selected Then, there is no need to occupy edge computing resources; it only needs to meet the corresponding communication resource conditions. .

[0163] When no feasible edge solution exists, the task is transferred to CSC for processing; when multiple feasible solutions exist, a greedy value is calculated based on three factors: economy, reliability, and load balancing.

[0164]

[0165] The economic factor in the greedy value measures the cost advantage of ES processing relative to CSC processing, the reliability factor uses the overall trust value of ES, and the load balancing factor is determined by the current remaining resources. With initial free resources The proportions are determined, and all three parties work together to avoid selecting low-reliability or high-load nodes based solely on short-term profits.

[0166] After selecting the solution with the largest greedy value, if that solution is an edge processing solution... The system deducts the corresponding remaining Elasticsearch resources. And mark the sub-band occupancy status. Then proceed to the next task.

[0167] Experimental verification: (I) Experimental Setup Simulation experiments were conducted to evaluate the performance of this invention using random task loads and preset attack behaviors. Each task consisted of triples. This indicates the amount of input data. Calculation workload and maximum tolerable delay They are generated randomly; forgery attacks are simulated by falsely reporting idle resources or shortening the reporting completion time, while conspiracy attacks are simulated by raising the trust evaluation between ES.

[0168] (ii) Baseline comparison To verify the effectiveness of the multi-source trust model, Lagrange multiplier allocation method, and trust-aware greedy offloading method proposed in this invention, the following methods were selected as comparative baselines. Each baseline is compared from the perspectives of search strategy, resource allocation strategy, integrity of trust modeling, and existing trust offloading frameworks.

[0169] Trusted Exhaustive Offloading Combined with LMAM (TEOJL): This method offloads all tasks to a candidate set of edge servers and performs an exhaustive search for feasible offloading combinations. Simultaneously, it uses the multi-source trust model from this invention to evaluate the trust level of the edge servers and allocates computing resources using LMAM. This method can serve as an approximate upper bound for performance analysis of the suboptimal nature of the greedy offloading strategy in this invention. However, its combinatorial search complexity increases rapidly with the number of end users and edge servers, making it difficult to meet the needs of large-scale real-time scheduling.

[0170] Trusted Random Offload Combined with LMAM (TROJL): End-user computing tasks are randomly assigned to edge servers. If an edge server cannot meet the constraints, the task is transferred to the cloud service center. Simultaneously, the multi-source trust model and LMAM resource allocation method of this invention are retained. This baseline is used to compare the differences between random offload and the trust-aware greedy offload of this invention, demonstrating the role of TGOM in candidate node selection, resource status updates, and load balancing.

[0171] Trusted Minimum Resource Allocation Combined with TGOM (TRAJT): This approach retains the multi-source trust model and TGOM offloading strategy from this invention, but allocates only the minimum computational resources that satisfy the maximum tolerable latency constraint to each task, without further optimizing the balance between benefits and costs through LMAM. This baseline is used to verify the improvement effect of the Lagrange multiplier allocation method on the system's net benefits and resource utilization efficiency.

[0172] Trustless MTCORA (NT-MTCORA): This baseline removes multi-source trust models, confidence fusion, and trust penalties during offloading and resource allocation, making decisions solely based on traditional factors such as cost, resources, and latency. It highlights the necessity of trust assessment and trust constraints in identifying dishonest edge servers, suppressing the participation of low-trust nodes, and mitigating the risk of task failure.

[0173] TCO Framework: TCO is a recently proposed trust-based computation offloading framework that employs a two-layer trust evaluation at both the cloud and user ends, and solves the offloading strategy through stochastic adaptive search and gradient descent methods. Compared to this invention, TCO's trust sources do not include collaborative trust between edge servers, and it lacks a dynamic update mechanism based on the forgetting curve. Trust factors are mainly used for gain calculation and resource discounting, and are not deeply integrated into the trust-aware net benefit objective. This baseline is used to verify the advantages of this invention in multi-source fusion, trust updating, and the depth of coupling between trust and offloading decisions.

[0174] (III) Net Benefit Analysis of Trust Perception The perceived net benefit of trust reflects the effectiveness of the offloading framework and the role of trust in overall benefits. Figure 2 The TANB value varies with the number of end users. As the number of EUs increases, resource competition intensifies, and the system needs to allocate limited edge computing resources among more tasks. Therefore, the trust-aware net benefit of each method generally declines. TEOJL obtains the globally optimal combination through exhaustive search, forming a performance upper bound. The method of this invention makes decisions sequentially according to task urgency and solves resource allocation precisely using Lagrange multipliers. The result is closest to TEOJL, indicating that this invention can achieve near-optimal offloading and resource allocation effects with lower complexity.

[0175] Figure 3 The diagram illustrates the variation of the TANB value under different numbers of edge servers. An increase in the number of Elasticsearch Nodes (ES) means more candidate offload nodes and available sub-bands, improving the freedom of resource matching. While random offload methods may also obtain more candidate nodes, they lack joint screening based on trust value, cost, and load status, resulting in limited benefit improvement. The method in this invention utilizes a comprehensive trust factor to guide tasks to prioritize high-trust ESs and uses a load balancing factor to avoid excessive resource concentration. Therefore, it can more fully utilize the resource gains brought by the newly added ESs, and the trust-aware net benefit is significantly higher than other comparative methods.

[0176] Figure 4This illustrates the variation of the TANB value under different idle computing resource conditions. The more abundant the idle resources of the ES (Executable Executor), the easier it is for tasks to obtain computing resources that meet latency constraints, thus increasing the trust-aware net benefit. When resources are scarce, the method of this invention prioritizes urgent tasks through minimum resource constraints and LMAM (Leadership Management Machine), and suppresses the participation of low-trust ESs through trust penalties. When resources increase, TGOM (Trusted Executable Machine) can make a joint choice between economy, reliability, and load balancing, translating the additional resources into higher net benefits. Therefore, Figure 4 This further demonstrates that the present invention has a stable advantage in both resource-constrained and resource-sufficient scenarios.

[0177] (iv) Trust convergence analysis Figure 5 This section describes the convergence of the overall trust value under a forgery attack scenario. The overall trust value of honest ES converges within a few cycles and then stabilizes near 1. For dishonest ES, their actions of falsely reporting resources and tampering with records cause a rapid decline. Compared to TCO, the trust evaluation value of the method in this invention better reflects the characteristics of dishonest behavior.

[0178] Figure 6 This demonstrates the response of different methods to dynamic changes in node behavior. When a dishonest ES essentially transforms into an honest ES, it rapidly ascends to a high trust level, proving that the method of this invention is highly sensitive to node behavior and possesses a rapid response capability.

[0179] Figure 7 This study examines the convergence of the comprehensive trust value under a conspiracy attack scenario. The method described in this invention exhibits the most rapid decrease and the lowest value, demonstrating faster convergence performance and higher identification capability compared to other methods. This result stems from the fact that the multi-source trust model of this invention employs a forgetting curve update mechanism, which rapidly decays historical negative records and allows recent true performance to dominate the trust value.

[0180] The foregoing description, with reference to preferred embodiments, details an exemplary implementation of the multi-source trust-driven computation offloading and resource allocation method and system proposed in this invention. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of this invention, and various combinations can be made to the various technical features and structures proposed in this invention without exceeding the protection scope of this invention, which is determined by the appended claims.

Claims

1. A multi-source trust-driven computation offloading and resource allocation method, characterized in that, The steps include the following: S10: The end user generates computing task information and sends it to the cloud service center. The computing task information includes the amount of input data, the required computing capacity, and the maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge server. S20: Calculate the trust value of the edge server from three dimensions: cloud service center, end user, and collaborative edge server, to obtain the cloud trust value. User trust and edge trust Three basic trust values; S30: Based on the confidence-weighted fusion method, the three types of basic trust values ​​are fused into a comprehensive trust value. ; S40: Employs an update mechanism based on the Ebbinghaus forgetting curve to update the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; S50: Construct a trust-perceived net benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation and computational offloading subproblems by incorporating a delay model and a cost function. The resource allocation subproblem is solved using the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task. The specific steps are as follows: S51: Set the task delay equal to the maximum tolerable delay. Find the minimum computing resources required while ensuring the latency requirement. Its expression is: In the formula, The computing capacity required to complete the task The overall trust value for edge servers; Size of the input data; The uplink transmission rate for users to upload tasks to the edge server; S52: Introduction of Lagrange multipliers The constrained optimization problem is transformed into an unconstrained Lagrangian function extremum problem, specifically expressed as: In the formula, Cost of computing resources per unit for edge servers; The amount of computing resources committed to the edge server for the task; This is the revenue conversion factor; The computing capacity required to complete the task; As a comprehensive trust value; Idle computing resources for edge servers; This represents the nth end user or its corresponding computing task that is offloaded to edge server m; Represents the Lagrange multiplier corresponding to the resource constraints of edge server m; S53: For each task, the candidate resource allocation value is obtained by ensuring that the partial derivatives are zero and that the complementary relaxation condition is satisfied. The expression is as follows: In the formula, Lagrange multipliers that satisfy the Karush-Kuhn-Tucker conditions; S54: Searching for the optimal Lagrange multiplier using the binary search method The following condition is true, and the specific expression is: In the formula, To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met; S60: Solve the computational unloading subproblem based on a trust-aware greedy unloading method, processing user-requested tasks sequentially according to their urgency, and using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

2. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, In step S20, cloud trust The calculation process is as follows: S21: Calculate the task timeout rate of the edge server. That is, the proportion of timed-out tasks to the total number of tasks; S22: Calculate service consistency by comparing task completion times in edge server and end-user records. Its expression is: In the formula, N is the length of the service record; For the i-th service time record of the end user; This is the i-th service time record for the edge server; S23: Obtain a cloud trust score by weighting task timeout rate and service consistency. Its expression is: In the formula, (▪) is the natural exponential function.

3. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, In step S20, edge trust is calculated. The calculation process is as follows: S201: When the primary edge server assigns a task to a collaborating edge server (ES) for collaborative processing, both parties record the collaborative response time. ; Calculate response timeliness Its expression is: In the formula, This is the time decay parameter; Among them, the time decay parameter is used to control the impact of collaborative response time on response timeliness, so that the longer the response time, the smaller the increment. S202: Calculate Collaboration Familiarity This refers to the number of collaborations provided to a specific edge server. The percentage of total collaborations is expressed as follows: In the formula, This refers to the number of collaborations provided to other edge servers; The number of edge servers that have collaborated; i is the iteration variable; S203: Edge Trust Score The weighted fusion of response timeliness and collaboration familiarity is expressed as follows: In the formula, Let represent the collaboration quality factor of the i-th collaborative edge server relative to edge server m.

4. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, In step S20, the user trusts... The calculation method is as follows: End users form trust ratings based on their direct interaction records with edge servers; After each interaction, the end user records the actual task completion time. Compared with the expected completion time Deviation; defining user-side trust Its expression is as follows: In the formula, The actual time to complete the task; This refers to the expected completion time of the task.

5. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, Step S30 specifically includes the following steps: S31: Treat the cloud service center, end users, and collaborative edge servers as different sources of trust; for each source of trust, the confidence level... The product of three factors: In the formula, For data volume factor, As a stability factor, Freshness factor; This represents the cumulative number of tasks corresponding to the trusted source. The variance of the trust value corresponding to the trust source; This refers to the latest update time corresponding to the trusted source; Among them, data volume factor Based on the number of interactions, the specific expression is: In the formula, To control parameters; regarding cloud trust, This refers to the cumulative number of tasks; regarding user trust, The number of end users participating in the evaluation; for edge trust, The number of edge servers that have collaborated; Among them, stability factor The variance calculation based on trust values ​​is specifically expressed as follows: In the formula, For control parameters; The variance of the trust value; Among them, freshness factor Calculated based on the time of the most recent update; In the formula, The current moment; This is the latest update time from the trusted source; This is the time decay parameter; S32: Overall Trust Value Defined as a weighted average of the confidence values, with the weights being the corresponding confidence levels: In the formula, The confidence level of a cloud trust source; Assess trust in the cloud; The confidence level of edge trust sources; Trust scoring at the edge; The number of edge servers that have collaborated; The confidence level of the end-user's trusted source; The number of end users served; Assess user trust rating; This represents the set of edge servers that have collaboration records with edge server m; Represents a set Any collaborative edge server in the network; This represents the set of end users served by edge server m; Represents a set The nth terminal user in the process.

6. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, In step S40, an update mechanism based on the Ebbinghaus forgetting curve is used to update the overall trust value. Dynamic updates involve the following steps: S41: Suppose that the edge server has gone through J evaluation cycles since the initial time, and obtains a comprehensive trust value at the end of each cycle. And record the time. For the first Trust value over a period of time, and its forgetting factor for: In the formula, The time interval between two updates; and These are the fitting parameters; S42: The current overall trust value is a weighted average of historical trust values. In the formula, For the first Period and the The time interval of each cycle.

7. The multi-source trust-driven computation offloading and resource allocation method according to claim 1, characterized in that, Step S60 specifically includes the following steps: S61: Generate all feasible ES solutions for each task, following trust conditions, subband conditions, transmission feasibility and resource feasibility constraints; S62: Sort tasks in descending order of urgency, where urgency is determined by the computation density of the task, and a higher urgency value indicates that the task should be processed first. S63: Calculate the greedy value for each feasible solution to the current task. : In the formula, Calculate the unit resource cost for the cloud service center; To determine the minimum computing resources required to offload tasks to the cloud service center while ensuring latency requirements are met; To determine the minimum computing resources required to offload tasks to edge servers while ensuring latency requirements are met; The remaining resources of the edge server at the current moment; Initial idle resources for edge servers; S64: Select the option with the largest greedy value as the unloading option for the current task; update the resource status and continue processing the next task.

8. A multi-source trust-driven computation offloading and resource allocation system, characterized in that, The computation offloading and resource allocation method for implementing the multi-source trust-driven method according to any one of claims 1-7 includes: The task information collection module is configured to generate computing task information for end users and send it to the cloud service center. This computing task information includes the amount of input data, required computing capacity, and maximum tolerable latency. The cloud service center communicates with each edge server to obtain the idle computing resources and unit computing resource cost of the edge servers. The multi-source trust assessment module is configured to calculate the trust value of the edge servers from three dimensions: the cloud service center, the end user, and the collaborating edge servers, respectively, to obtain the cloud-based trust value. User trust and edge trust Three basic trust values; The comprehensive trust fusion module is configured to fuse the three types of basic trust values ​​into a comprehensive trust value based on a confidence-weighted fusion method. ; The trust update module is configured to use an update mechanism based on the Ebbinghaus forgetting curve to update the overall trust value. Dynamically update the weight of historical trust values. With time interval Increase and decrease; The resource allocation module is configured to construct the Trust-Aware Net Benefit (TANB) maximization optimization problem, which will integrate trust values. The optimization problem is decomposed into resource allocation subproblems and computation unloading subproblems by incorporating the delay model and cost function; the resource allocation subproblem is solved based on the Lagrange multiplier allocation method to determine the optimal computational resource allocation scheme for each task; The computational unloading module is configured to solve the computational unloading subproblem based on a trust-aware greedy unloading method. It processes user-requested tasks sequentially according to their urgency, using a greedy value... When selecting a high-quality unloading solution, the greedy value takes into account three factors: economy, reliability, and load balancing.

9. The multi-source trust-driven computation offloading and resource allocation system according to claim 8, characterized in that, The multi-source trust assessment module includes: Cloud-based trusted computing unit, configured for task timeout rate based on edge server Service Consistency Calculate cloud trust scores ; The edge-side trust computing unit is configured to provide timely responses based on collaborative edge servers. and familiarity with collaboration Calculate edge trust score ; The user-side trust calculation unit is configured to base its calculations on the actual task completion time recorded in the interaction logs between the end user and the edge server. Compared with the expected completion time Deviation calculation of user trust score .

Images