CN113641500A - Offshore edge calculation unloading method for comprehensive trust evaluation - Google Patents

Abstract

The invention discloses a maritime edge computing unloading method for comprehensive trust evaluation, which comprises the steps of establishing a node trust forgetting function, constructing an inter-node unloading request and a cooperation evaluation model by taking task authenticity and node unloading cooperation behavior characteristics as constraints, identifying and filtering false tasks and bad cooperation nodes in a network, designing a search discovery probability optimization particle swarm algorithm, avoiding node overload caused by node reuse, improving unloading efficiency of edge computing tasks and effective utilization of resources, and keeping network robustness; simulation experiments show that by constructing the unloading cooperative service system of comprehensive trust evaluation, the task execution success rate of the offshore edge computing network can be effectively improved, the resource loss is reduced, the safe cooperation, load balance and effective resource utilization of offshore edge computing are realized, and the unloading comprehensive utility is improved.

Description

Offshore edge calculation unloading method for comprehensive trust evaluation

Technical Field

The invention relates to the technical field of offshore edge computing, in particular to an offshore edge computing unloading method for comprehensive trust evaluation.

Background

The rapid development of the intelligent internet creates a new opportunity for maritime activities, along with the popularization of intelligent terminal application and data-driven maritime services, a large number of intelligent terminal devices are deployed in a maritime network, the maritime communication requirements of ultra-reliability and low cost are increased, the edge computing which is required by maritime activities and cannot be met by a processing mode of using cloud computing by satellite relays is taken as a novel computing mode, an edge server is deployed between the maritime intelligent terminal devices and a land data center, computing and storage resources are brought to the edge side of the maritime network, low delay, low energy consumption, privacy protection and the like are achieved, and the network QoS and the comprehensive utility are improved; the method has the advantages that the intelligent terminal with resource limitation is effectively fused and dispatched, the requirement for meeting huge and diversified computing requirements is important content of offshore edge computing task unloading research, task unloading is used as one of key technologies of edge computing cooperative services, the three problems of unloading, task dividing and resource allocation of tasks in edge computing are mainly solved, the computing tasks are unloaded from equipment with resource limitation to equipment or networks with rich resources, QoE of users is improved, expenses are reduced, and fusion and resource sharing are achieved.

The maritime network environment is complex and changeable, the equipment types are complex, diverse and heterogeneous, and the method has the characteristics of dynamics, complexity, unreliability and the like, due to the ubiquitous existence of bad nodes such as selfish, rationality and malice, a task unloading service requester tends to optimize an unloading task, a non-cooperative node task unloading service requester continuously issues a false task, and a selfish and rational resource provider seizes and consumes resources such as calculation, storage and energy in a network for ensuring that a self-interest taking a bus and a malice resource provider starts On-off attack and other bad behaviors for intercepting the task, and partial nodes are over-consumed to cause 'death' due to the excessive consumption of the resources, so that the task accumulation and execution failure are caused, the network throughput rate and the QoS are reduced, and the efficiency of the maritime edge calculation cooperative service is directly influenced.

At present, task offloading technologies have achieved many research achievements in performance optimization such as delay and energy consumption, but few research is made on problems such as whether task offloading service requesters and resource providers in collaborative services are true and reliable, and edge computing task offloading constrained by true reliability of task requesters and resource providers is a very challenging problem.

Disclosure of Invention

Aiming at the problems that under the offshore Edge Computing environment, Edge Computing equipment is complex and diverse in type and heterogeneous, nodes are restricted by factors such as dynamics, complexity and unreliability, a large number of false tasks are faced during task unloading, the capability of node unloading cooperative service is poor, the effective utilization rate of resources is low, part of nodes are overloaded and crashed, the unloading service quality is poor and the like, the invention provides an offshore Edge Computing unloading model (MECO-CTE) with Comprehensive Trust Evaluation, establishes a node Trust forgetting function and a node behavior punishing operator based on the time relevance of node behaviors, establishes a cooperative Trust Evaluation mechanism based on task unloading requests and node unloading, inhibits the bad behaviors, avoids invalid time and resource expenses caused by repeated task unloading due to false tasks and node jitter, and realizes reliable unloading service of Computing tasks, simulation results show that the system model can effectively identify false tasks of non-cooperative nodes in a filtering network, provide unloading cooperative service for selecting the optimal cooperative node for the task, and realize safe filtering and reliable cooperation of edge calculation;

the technical scheme for realizing the purpose of the invention is as follows:

a maritime edge computing unloading method based on comprehensive trust evaluation realizes task unloading cooperative service through a task filtering and unloading service decision mechanism, establishes a node trust forgetting function and a node behavior reward and punishment operator, establishes a task filtering based on task unloading request behavior evaluation and an unloading service decision mechanism based on unloading cooperative comprehensive evaluation, and realizes false task filtering and reliable unloading service of a maritime network;

the task unloading cooperative service system takes a land data center as a cloud server, a base station and a floating platform as Super Nodes (SN), and an offshore naval vessel and terminal intelligent equipment as Edge nodes (Edge nodes, EN); locally aggregating EN nodes by taking the SN as a center to construct an edge computing cluster, realizing edge computing resource sharing and providing task unloading cooperative service;

in the task unloading cooperative service system, an SN node constructs an inter-node unloading request and a cooperative trust evaluation model according to task authenticity and node unloading cooperative behavior characteristics as constraints, and identifies and filters false tasks and bad cooperative nodes in a network;

in the task unloading cooperative service system, an SN node unloads cooperative trust according to cooperative candidate nodes, selects the cooperative nodes to construct a cooperative candidate node set, designs a Particle swarm optimization (PS-SDP) algorithm based on search and discovery probability improvement to solve the unloading cooperative trust maximization problem, and realizes cooperative node load balancing and efficient unloading;

the offshore edge computing unloading method for comprehensive trust evaluation comprises the following steps:

1) a source node initiates a task unloading request to an SN node;

2) the SN node constructs an unloading request behavior evaluation model according to task authenticity, calculates unloading request trust of a source node, judges node cooperation and filters false tasks of non-cooperative nodes;

3) the SN node evaluates the resource trust and the unloading behavior trust of the EN node according to the parameters of the EN node, such as the calculation storage capacity, the task execution success rate, the execution efficiency and the like, and constructs an unloading cooperation comprehensive evaluation model; setting a trust threshold value, and selecting a trusted EN node to construct an unloading cooperation candidate node set; designing a PS-SDP algorithm to solve the optimal unloading cooperation EN node by taking the unloading cooperation trust maximization as a target;

4) the SN node sends an unloading request to the EN node; the EN node autonomously selects to accept or refuse to respond to the task unloading request from the SN node, and if the node accepts the request, the EN node executes and completes the task unloading service; otherwise, the SN node marks the EN node, and searches the cooperative node again by using the PS-SDP algorithm;

5) the EN node returns the execution result of the task and feeds back the real characteristic information of the task to the SN node; updating node behavior and task characteristic information;

6) updating the SN node unloading request trust level according to the task real characteristic information fed back by the EN node; recalculating the node behavior trust according to the unloading behavior characteristics of the current EN node; and judging whether the behavior trust of the node is lower than a neutral value, if so, generating a group of test task sets to be executed by the node, implementing excitation, and calculating and updating the behavior trust of the node again.

In step 2), the SN node constructs an unloading request behavior evaluation model according to task authenticity, calculates the unloading request trust of a source node, judges node cooperation, and filters false tasks of non-cooperative nodes, and specifically comprises the following steps:

2-1) setting initial trust degree and trust threshold of SN node i unloading request trust

2-2) constructing an unloading request behavior evaluation model, and judging an unloading request trust relationship between the unloading request behavior evaluation model and a request node:

task unloading is a dynamic decision process with time relevance, and the trust characteristics of the edge computing nodes have time attenuation attributes; constructing a decay function f (t) of the trusted memory retention of the edge computing node with respect to time t based on Ebbinghaus human brain memory forgetting theory:

dividing each observation period into t_kConstructing a historical trust forgetting function of the edge computing node in each time slice as follows:

wherein T(s) represents the trust of the s-th observation period, Δ s represents the observation period interval number, σ represents a forgetting factor, and λ represents the retention rate of the historical trust;

let the number of real tasks and dummy tasks unloaded by node j in edge computing cluster i (edge computing cluster with SN node i as center) during s observation period be

Calculating the task real rate of the SN node j according to the task execution condition in the cluster

Comprises the following steps:

in order to inhibit adverse behaviors of non-cooperative node manufacturing, unloading false tasks and the like, improve trust confidence coefficient of cooperative nodes and construct reward and punishment operators of node unloading request behaviors

Introducing a history trust forgetting function, and calculating the unloading request trust of the node j after s observation periods

Wherein the content of the first and second substances,

a reward and penalty operator representing the behavior of the offload request,

representing the actual task trust of the node i to the node j during the s observation period;

2-3) based on offload request trust threshold

Judging the cooperative relationship between the SN node i and the task unloading source node j; if it is

The node j is a non-cooperative node, the node i rejects the task unloading request of the node j, and 2-4 is switched to), otherwise, the task unloading service is executed;

2-4) starting an unloading service decision mechanism, selecting an optimal cooperative node, executing a task by the cooperative node and judging the authenticity of the task, and updating real characteristic information of the task by the SN node according to a feedback result of the cooperative node;

and 2-5) if the observation period is over, updating the unloading request trust of each SN node according to the real characteristic information of the task and by combining the historical trust relationship.

In step 3), the SN node evaluates the resource trust and the unloading behavior trust thereof, constructs an unloading cooperation comprehensive evaluation model, selects a trusted EN node to construct an unloading cooperation candidate node set, designs a PS-SDP algorithm to solve an optimal unloading cooperation EN node, and specifically comprises the following steps:

3-1) setting initial trust degree and trust threshold of SN node i for unloading cooperative trust

3-2) updating the resource trust T according to the residual resources of the EN node_i ^resFederated behavioral trust

Compute offload collaboration trust

The specific modeling process for offloading the collaboration trust is as follows:

3-2-1) computing node resource trust T based on EN node residual computing resources and storage resources_i ^res：

The maximum load rate of the node is set as rho, and the given node i can provide computing resources and storage resources which are respectively C_i，M_iWhen unloading task r, node i resource trust degree T_i ^resComprises the following steps:

wherein, c_iAnd m_iRepresenting the current remaining computing and storage resources of node i, c_rAnd m_rRepresents the computational and memory resources required for task r execution;

in order to avoid the influence of congestion, jitter and the like on the network stability, the unloading and the execution of tasks are completed in the same observation period, and the online time of the node i is set as

After task r is executed, its remaining online time

Comprises the following steps:

the EN node completes the computational task while remaining online during the observation period, with offloading limited to:

wherein, t_cycThe length of the observation period is indicated,

which represents the computing power of the node i,

representing the elapsed time before the task is executed during the observation period;

3-2-2) SN node takes task execution efficiency and success rate of node as constraints to construct node behavior trust model

Task execution efficiency of EN node j in edge computing cluster

The calculation formula is as follows:

the number of tasks received by the node j in the cluster in the s observation period is known as

The number of tasks accepted and successfully executed is

Task execution success rate

The calculation formula is as follows:

in order to inhibit the adverse behaviors of EN nodes in the cluster, stimulate the node integrity service, and set the reward and punishment operator of the node unloading cooperative behavior

Calculating task execution trust T of node i to node j in s observation period^suc(s)：

Wherein, ω represents the weight occupied by the task execution success rate,

representing the number of tasks unloaded from the SN node i to the node j in the s observation period;

combining node task execution efficiency

And execution trust T^suc(s) calculating the behavioral trust of the node j in the s observation period

Introducing a history trust forgetting function, and calculating the behavior trust of the node j after s observation periods

3-2-3) incorporate node resource trust T_i ^resAnd behavioral trust

Building an offload collaboration trust model

3-3) obtaining EN node unloading cooperative trust by the formula (15)

Offloading cooperative trust thresholds from SN nodes

Determining the node trust relationship if

The EN node j cooperative service is credible, and the unloading cooperative candidate node set is obtained by screening

3-4) converting the optimization problem of the unloading cooperative nodes into a target optimization problem of the unloading cooperative trust maximization, and collecting nodes

Designing a PS-SDP algorithm for solving a solution space, and constructing an unloading cooperative node optimizing search target and constraint conditions of a task r in an SN node i:

performing optimization search on cooperative nodes by using a particle swarm optimization algorithm, wherein the flight speed v of a particle l in a search space_lAnd position

The update formula is as follows:

wherein v is_l(τ) and x_l(τ) represents the velocity and candidate nodes of the particle l after τ iterations,

and x^gb(τ) represents the optimal trust nodes for particle l and the entire population of particles after τ iterations, c₁And c₂Learning factor representing the particles themselves and populations, w represents an inertia factor, w_maxAnd w_minRespectively representing maximum and minimum thresholds of the inertia factors, and G is the maximum iteration number of the algorithm; introducing search finding probability p E [0,1 ∈ ]]And when a number from 0 to 1 is randomly generated and is smaller than the finding probability, discarding the current solution and searching a new solution again.

In step 3-4), the PS-SDP algorithm solves the problem of the maximum uninstalling collaboration trust, and the specific steps are as follows:

3-4-1) setting the maximum rejection times of the tasks;

3-4-2) judging whether the rejected times of task unloading reach the maximum value, if so, directly unloading the task to other edge computing clusters; otherwise, initializing node optimization parameters including particle number, iteration number tau, maximum iteration number and optimization speed v_l(τ) maximum and minimum values, learning factor c₁And c₂Maximum value of inertia factor w_maxAnd a minimum value w_minEach particle represents a potential offload collaboration candidate node; in offloading collaborative candidate node sets

Computing particle candidate node offload cooperative trust according to equation (15)

Obtaining an initial particle swarm;

3-4-3) judging whether the tau reaches the maximum iteration times; if the maximum iteration times are not reached, updating the particle flight speed and the candidate nodes according to the formula (17) to generate a new particle swarm; carrying out unloading cooperation trust evaluation on the candidate node of each particle of the new particle swarm, and searching a new optimal unloading cooperation trust node according to the formula (16); judging whether the probability p is found in accordance with the search, if so, randomly updating the particle candidate nodes to generate a new particle swarm, performing unloading cooperative trust evaluation on each particle candidate node, and updating the optimal unloading cooperative trust node; otherwise, entering the next iteration, wherein tau is tau + 1;

3-4-4) stopping iterative computation when the iteration times tau reach the maximum value to obtain the best trust node of the group; taking the node as a cooperative node and initiating an unloading request;

3-4-5) judging whether the cooperative node refuses the unloading request, if so, turning to 3-4-2), and searching a new cooperative node again; otherwise, performing task unloading service.

Has the advantages that: the invention provides an offshore edge computing unloading method for comprehensive trust evaluation, which aims to solve the problems of malicious resource invasion and resource consumption of bad behaviors such as false cheating, vehicle taking and the like in an edge computing environment by an edge computing unloading model integrating false task filtering and bad cooperative node identification functions, establish a node trust forgetting function, construct an unloading request and a cooperative evaluation model among nodes by taking resource allowance and interactive behaviors as constraints, identify and filter false tasks and bad cooperative nodes in a network, provide a mathematical model and constraint conditions for constructing false task filtering and unloading cooperative services, and analyze the mathematical model and the constraint conditions in more detail; simulation experiments show that the execution success rate of the real tasks of the MECO-CTE strategy model reaches 59.22%, the resource loss rate reaches 6.35% at least, compared with a random unloading strategy based on an on-demand routing protocol, the execution success rate of the real tasks of the MECO-CTE strategy model is improved by 30.36%, the resource loss rate is reduced by 5.23%, false tasks can be effectively inhibited and filtered, and the reliable collaborative service efficiency of offshore edge calculation is improved.

Drawings

FIG. 1 is a system offloading collaborative services process;

FIG. 2 is a task offload collaboration services system trust framework;

FIG. 3 is a task filtering mechanism based on offload request behavior evaluation;

FIG. 4 is an offload service decision based on offload collaboration composite evaluation;

FIG. 5 is a flow of cooperative node optimization within a cluster;

FIG. 6 is a model of a marine edge computing network;

FIG. 7 is a graph showing a memory retention decay curve;

FIG. 8 is a graph illustrating resource utilization variation for different offload cooperative service systems;

FIG. 9 is a graph of non-cooperative node identification rate and identification accuracy rate changes for different offload request trust thresholds;

FIG. 10 is a graph of variation in offload request rejection rate and rejection accuracy for different offload request trust thresholds;

FIG. 11 is a graph illustrating the change in effective utilization of resources for different offload request trust thresholds;

FIG. 12 is a graph of non-cooperative node identification rate and identification accuracy rate variation for different network scales;

FIG. 13 is a graph of variation in denial of offload requests and denial of accuracy across different network scales;

FIG. 14 is a graph of resource availability for different network sizes;

FIG. 15 is a graph illustrating variation of online time lengths of edge computing clusters under different offloading cooperative policies

FIG. 16 is a graph illustrating a variation of a task offload rate under different offload coordination strategies;

FIG. 17 is a diagram illustrating a variation of task failure rates under different offloading coordination policies;

FIG. 18 is a diagram illustrating a variation of task denial times under different offloading coordination policies;

FIG. 19 is a diagram illustrating trust change of behavior of an incentive node under different cooperative node optimization algorithms in an offload cooperative service system with trusted cooperative nodes;

FIG. 20 is a diagram illustrating behavior trust changes of reliable nodes under different cooperative node optimization algorithms in an offload cooperative service system with trusted cooperative nodes;

FIG. 21 is a diagram illustrating a change in the failure rate of task execution under different cooperative node optimization algorithms in an offloaded cooperative service system in which a cooperative node is trusted;

FIG. 22 is a diagram illustrating a variation of a success rate of executing a real task in different task offloading coordination service systems;

FIG. 23 is a graph illustrating resource consumption rate changes of edge computing clusters of different task offloading cooperative service systems.

Detailed Description

The invention will be further described with reference to the following drawings and examples, but the invention is not limited thereto;

example (b):

assuming a marine edge computing scene, taking a land data center as a cloud server, a base station and a floating platform as super nodes SN, and a marine vessel and a terminal intelligent device as edge nodes EN, as shown in fig. 6, the SN scale is X ^sn1,2, m, EN scale is

The SN nodes and the root nodes construct an offshore cloud edge cooperative backbone network, the SN nodes and the EN nodes construct an offshore edge computing network, and the EN nodes are locally aggregated by taking the SN as a center to construct an edge computing cluster, so that edge computing resource sharing is realized, and task unloading cooperative service is provided; suppose that:

1) 30% of SN nodes are non-cooperative nodes, the initial trust degree of the unloading request is 1, the SN nodes respond to 30 times of unloading service requests from other edge computing clusters in a unit time slice at most and only accept one unloading request, and the number of tasks contained in each unloading request follows Poisson distribution; the time overhead generated by decision time intervals between adjacent tasks in the same time slice and unloading negotiation between nodes is ignored;

2) in the edge computing cluster, EN nodes are divided into three types, namely reliable nodes, excitation nodes and bad nodes: (1) and (3) reliable node: the unloading cooperation behavior is credible, and the success rate is higher; (2) exciting the node: in the initial stage, the unloading cooperative service capability is poor, and the nodes of the service enthusiasm are mobilized by excitation; (3) and (3) bad nodes: nodes that provide collaboration services in a passive service attitude for a long period of time; the number ratio of reliable, exciting and bad EN nodes is 6:2: 2; the initial unloading cooperation trust degree of the node is 0.5, and the tasks of the downlink unloading can be accumulated in the EN node, but are all completed in the observation period and are not retained in the next period;

constructing a maritime edge computing unloading cooperative service system, as shown in fig. 1, and implementing the steps as follows:

1) initializing a system, wherein in offshore edge calculation, SN node aggregation EN nodes construct independent and autonomous edge calculation clusters, each cluster comprises 1 SN node, and the cluster scales are kept consistent;

2) a false task filtering mechanism based on unloading request trust evaluation is established in a communication link for unloading tasks among SN nodes, false tasks of non-cooperative nodes are identified and filtered, and delay overhead and resource loss caused by false task unloading of bad nodes in offshore edge calculation are suppressed;

3) a task unloading cooperation service mechanism based on unloading cooperation comprehensive trust evaluation is set on a communication link for task unloading of the SN node and the EN node in the edge computing cluster, so that reliable unloading of computing tasks is guaranteed;

4) suppose a node randomly initiates a task unloading request, and the example is divided into 7 groups:

random Walk based Random Walk Random offload of collaborative services systems: the SN node directly receives the task unloading request and selects the EN node to execute the unloading service in a random walk mode;

an On-Demand Routing protocol (Ad hoc On-Demand Distance Vector Routing, AODV) -based random offload collaboration service system: the SN node directly receives the task unloading request and randomly selects the EN node to execute the unloading service based on the AODV;

task Optimal Offloading (TOO) collaborative services system: the SN node directly receives the task unloading request, and the optimization-oriented EN node executes unloading service;

task trusted offload (TCO) collaboration services system: the SN node evaluates the unloading request behavior of the source node by using an unloading request behavior evaluation model, filters false tasks and randomly selects an EN node to execute unloading service;

task optimization and Offloading (TOO-TC) collaborative service system based on Task credibility: the SN node evaluates the unloading request behavior of the task unloading source node by using an unloading request behavior evaluation model, filters false tasks, and prefers to select the EN node to execute unloading service;

a Cooperative Node trusted offload (CNCO) Cooperative service system: the SN node directly receives a task unloading request, an EN node is evaluated by using an unloading cooperation comprehensive evaluation model, and an optimal cooperative node is selected to execute unloading service;

task offload (MECO-CTE) collaborative services system based on Comprehensive Trust Evaluation: the SN node evaluates the unloading request behavior of the task unloading source node by using an unloading request behavior evaluation model, and filters false tasks; evaluating an EN node by using an unloading cooperation comprehensive evaluation model, and selecting an optimal cooperation node to execute unloading service;

the configuration parameters of the cooperative service system are shown in the table 1;

specifically implementing the offshore edge computing offloading method of comprehensive trust evaluation, as shown in fig. 2, constructing an offloading request behavior evaluation model, and filtering and suppressing false offloading task requests of non-cooperative nodes; the method comprises the following steps of constructing an unloading cooperation comprehensive evaluation model, selecting an optimal cooperation node, unloading a calculation task by a SN node in a pair of multi-mode to obtain high-efficiency reliable unloading service, and realizing unloading cooperation service optimization of an edge calculation task, wherein the method comprises the following steps:

1) a source node initiates a task unloading request to an SN node;

2) the SN node constructs an unloading request behavior evaluation model according to task authenticity, calculates unloading request trust of a source node, judges node cooperation, and filters false tasks of non-cooperative nodes, as shown in FIG. 3, the specific steps are as follows:

2-1) setting initial trust degree and trust threshold of SN node i unloading request trust

2-2) constructing an unloading request behavior evaluation model, and judging an unloading request trust relationship between the unloading request behavior evaluation model and a request node:

constructing a decay function f (t) of the trusted memory retention of the edge computing node with respect to time t based on Ebbinghaus human brain memory forgetting theory:

wherein, the value α is 0.3574, β is 531.7, and the memory attenuation curve is shown in fig. 7; dividing each observation period into t_kConstructing a historical trust forgetting function of the edge computing node in each time slice as follows:

wherein T(s) represents the trust of the s-th observation period, Δ s represents the observation period interval number, σ represents a forgetting factor, and λ represents the retention rate of the historical trust;

let the number of real tasks and dummy tasks unloaded by node j in edge computing cluster i (edge computing cluster with SN node i as center) during s observation period be

Calculating the task real rate of the SN node j according to the task execution condition in the cluster

Comprises the following steps:

in order to inhibit adverse behaviors of non-cooperative node manufacturing, unloading false tasks and the like, improve trust confidence coefficient of cooperative nodes and construct reward and punishment operators of node unloading request behaviors

Introducing a history trust forgetting function, and calculating the unloading request trust of the node j after s observation periods

Wherein the content of the first and second substances,

a reward and penalty operator representing the behavior of the offload request,

representing the actual task trust of the node i to the node j during the s observation period;

2-3) based on offload request trust threshold

Judging the cooperative relationship between the SN node i and the task unloading source node j; if it is

The node j is a non-cooperative node, the node i rejects the task unloading request of the node j, and 2-4 is switched to), otherwise, the task unloading service is executed;

2-4) starting an unloading service decision mechanism, selecting an optimal cooperative node, executing a task by the cooperative node and judging the authenticity of the task, and updating real characteristic information of the task by the SN node according to a feedback result of the cooperative node;

2-5) if the observation period is over, updating the unloading request trust of each SN node according to the real characteristic information of the task and by combining the historical trust relationship;

3) the SN node evaluates the resource trust T of the EN node according to the parameters of the EN node, such as calculation storage capacity, task execution success rate, execution efficiency and the like_i ^resAnd offload behavior messagesRen

Construction of unloading cooperation comprehensive evaluation model

As shown in FIG. 4, a trust threshold is set

Selecting a credible EN node to construct an unloading cooperation candidate node set; with the maximization of the uninstalled cooperation trust as a target, designing a PS-SDP algorithm to solve the optimal uninstalled cooperation EN node, as shown in FIG. 5, the specific steps are as follows:

3-1) setting initial trust degree and trust threshold of SN node i for unloading cooperative trust

3-2) updating the resource trust T according to the residual resources of the EN node_i ^resFederated behavioral trust

Compute offload collaboration trust

The specific modeling process for offloading the collaboration trust is as follows:

3-2-1) computing node resource trust T based on EN node residual computing resources and storage resources_i ^res：

The maximum load rate of the node is set as rho, and the given node i can provide computing resources and storage resources which are respectively C_i，M_iWhen unloading task r, node i resource trust degree T_i ^resComprises the following steps:

wherein, c_iAnd m_iRepresenting the current residual computation and storage of node iResource, c_rAnd m_rRepresents the computational and memory resources required for task r execution;

in order to avoid the influence of congestion, jitter and the like on the network stability, the unloading and the execution of tasks are completed in the same observation period, and the online time of the node i is set as

After task r is executed, its remaining online time

Comprises the following steps:

the EN node completes the computational task while remaining online during the observation period, with offloading limited to:

wherein, t_cycThe length of the observation period is indicated,

which represents the computing power of the node i,

representing the elapsed time before the task is executed during the observation period;

3-2-2) SN node takes task execution efficiency and success rate of node as constraints to construct node behavior trust model

Task execution efficiency of EN node j in edge computing cluster

The calculation formula is as follows:

the number of tasks received by the node j in the cluster in the s observation period is known as

The number of tasks accepted and successfully executed is

Task execution success rate

The calculation formula is as follows:

in order to inhibit the adverse behaviors of EN nodes in the cluster, stimulate the node integrity service, and set the reward and punishment operator of the node unloading cooperative behavior

Calculating task execution trust T of node i to node j in s observation period^suc(s)：

Wherein, ω represents the weight occupied by the task execution success rate,

representing the number of tasks unloaded from the SN node i to the node j in the s observation period;

combining node task execution efficiency

And execution trust T^suc(s) calculating the behavioral trust of the node j in the s observation period

Introducing a history trust forgetting function, and calculating the behavior trust of the node j after s observation periods

3-2-3) incorporate node resource trust T_i ^resAnd behavioral trust

Building an offload collaboration trust model

3-3) obtaining EN node unloading cooperative trust by the formula (15)

Offloading cooperative trust thresholds from SN nodes

Determining the node trust relationship if

The EN node j cooperative service is credible, and the unloading cooperative candidate node set is obtained by screening

3-4) converting the optimization problem of the unloading cooperative nodes into a target optimization problem of the unloading cooperative trust maximization, and collecting nodes

Designing a PS-SDP algorithm for solving a solution space, and constructing an unloading cooperative node optimizing search target and constraint conditions of a task r in an SN node i:

performing optimization search on cooperative nodes by using a particle swarm optimization algorithm, wherein the flight speed v of a particle l in a search space_lAnd position

The update formula is as follows:

wherein v is_l(τ) and x_l(τ) represents the velocity and candidate nodes of the particle l after τ iterations,

and x^gb(τ) represents the optimal trust nodes for particle l and the entire population of particles after τ iterations, c₁And c₂Learning factor representing the particles themselves and populations, w represents an inertia factor, w_maxAnd w_minRespectively representing maximum and minimum thresholds of the inertia factors, and G is the maximum iteration number of the algorithm; introducing search finding probability p E [0,1 ∈ ]]When a number from 0 to 1 is randomly generated and is smaller than the discovery probability, discarding the current solution and searching a new solution again;

in step 3-4), the PS-SDP algorithm solves the problem of the maximum uninstalling collaboration trust, and the specific steps are as follows:

3-4-1) setting the maximum rejection times of the tasks;

3-4-2) judging whether the rejected times of task unloading reach the maximum value, if so, directly unloading the task to other edge computing clusters; otherwise, initializing node optimization parameters (the particle number is 10, the iteration number tau is 0, the maximum iteration number is 6, and the optimization speed v is_lMaximum and minimum values of (tau) of 5 and-5, learning factor c₁And c₂0.5, maximum value w of inertia factor_maxAnd a minimum value w_min0.8, 0.4, respectively), each particle represents a potential offload cooperation candidate node; in offloading collaborative candidate node sets

Computing particle candidate node offload cooperative trust according to equation (15)

Obtaining an initial particle swarm;

3-4-3) judging whether the tau reaches the maximum iteration times; if the maximum iteration times are not reached, updating the particle flight speed and the candidate nodes according to the formula (17) to generate a new particle swarm; carrying out unloading cooperation trust evaluation on the candidate node of each particle of the new particle swarm, and searching a new optimal unloading cooperation trust node according to the formula (16); judging whether the probability p is found in accordance with the search, if so, randomly updating the particle candidate nodes to generate a new particle swarm, performing unloading cooperative trust evaluation on each particle candidate node, and updating the optimal unloading cooperative trust node; otherwise, entering the next iteration, wherein tau is tau + 1;

3-4-4) stopping iterative computation when the iteration times tau reach the maximum value to obtain the best trust node of the group; taking the node as a cooperative node and initiating an unloading request;

3-4-5) judging whether the cooperative node refuses the unloading request, if so, turning to 3-4-2), and searching a new cooperative node again; otherwise, performing task unloading service;

4) the SN node sends an unloading request to the EN node; the EN node autonomously selects to accept or refuse to respond to the task unloading request from the SN node, and if the node accepts the request, the EN node executes and completes the task unloading service; otherwise, the SN node marks the EN node, and searches the cooperative node again by using the PS-SDP algorithm;

5) the EN node returns the execution result of the task and feeds back the real characteristic information of the task to the SN node; updating node behavior and task characteristic information;

6) updating the SN node unloading request trust level according to the task real characteristic information fed back by the EN node; recalculating the node behavior trust according to the unloading behavior characteristics of the current EN node; and judging whether the behavior trust of the node is lower than a neutral value, if so, generating a group of test task sets to be executed by the node, implementing excitation, and calculating and updating the behavior trust of the node again.

Aiming at the step 5), the evaluation scheme for constructing the offshore edge computing unloading method of the comprehensive trust evaluation is as follows:

5-1) aiming at task unloading among edge computing clusters, verifying effectiveness of an MECO-CTE model on filtering false tasks and improving resource utilization by taking non-cooperative node identification rate, identification accuracy rate, unloading request rejection rate, rejection accuracy rate, resource effective utilization rate and the like as indexes, wherein the evaluation indexes are as follows:

5-1-1) non-cooperative node identification rate

Rate of non-cooperative node identification during observation period

The ratio of the number of non-cooperative SN nodes identified for SN node i to the total number of SN nodes in the edge calculation:

wherein the content of the first and second substances,

representing non-cooperative SNs identified by SN node i in edge computationA set of nodes is provided, wherein,

representing the cooperativity of SN node j identified by node i, if node j is identified as a cooperative node

On the contrary, the method can be used for carrying out the following steps,

5-1-2) accuracy of non-cooperative node identification

Accuracy rate of non-cooperative node identification in observation period

The ratio of the number of nodes without error in identity to the total number of nodes in the non-cooperative nodes identified by the SN node i is as follows:

wherein the content of the first and second substances,

representing the actual cooperation of the SN node j in the edge calculation, if the node is actually a cooperative node

On the contrary, the method can be used for carrying out the following steps,

5-1-3) offload request rejection Rate

In the MECO-CTE model, if the SN node is judged to be a non-cooperative node, all the calculation tasks issued by the node are considered as false tasks, and the unloading request is rejected; rate of rejection of offload requests during observation period

Is SN nodeThe ratio of the number of tasks rejected by the point i to the number of tasks requested to unload received:

wherein the content of the first and second substances,

indicating the number of tasks from the SN node j in the unloading request received by the SN node i;

5-1-4) offload request rejection accuracy

Rate of accuracy of rejection of offload requests during observation period

The ratio of the number of false tasks to the total number of tasks rejected by the SN node i is as follows:

5-1-5) effective utilization of resources

In the edge computing cluster, the effective utilization rate of resources is measured by the real rate of task execution; effective utilization rate of resources in observation period

For the SN node i accepts tasks, the ratio of the number of tasks from the cooperative node to the total number of tasks:

as can be seen from the above equation, the effective utilization rate of the resource and the number of the identified non-cooperative nodes are in an inverse correlation relationship.

5-2) aiming at the intra-cluster unloading of the tasks, verifying the unloading service reliability and load controllability of the MECO-CTE model by taking the online time, the unloading rate, the task execution failure rate, the reject times, the unloading decision time and the like of an edge computing cluster as indexes, wherein the evaluation indexes are as follows:

5-2-1) online time duration: in order to avoid extra overhead generated by downtime of EN node equipment in a cluster, the online time of each node is prolonged as much as possible; length of time on line in observation period

The remaining time of the shortest EN node in the cluster is as follows:

wherein the content of the first and second substances,

representing the remaining online time of the EN node j at the end of the observation period;

5-2-2) task unloading rate: rate of task offloading during observation period

Calculating the ratio of the number of tasks unloaded to other clusters to the total number of tasks in the tasks received by the cluster i for the edge:

wherein the content of the first and second substances,

indicating the number of tasks offloaded to other clusters;

5-2-3) failure rate of task execution: failure rate of task execution in observation period

For the ratio of the number of the tasks which are accepted and executed by the edge computing cluster i and fail to be executed to the total number of the tasks:

5-2-4) number of task rejections: the time overhead generated by the unloading negotiation between the nodes is measured by the task rejection times, and the more the task rejection times are, the larger the time overhead generated by the unloading negotiation is; number of task rejections within observation period

And (3) for the sum of times that each EN node rejects task unloading requests in the edge computing cluster i:

wherein the content of the first and second substances,

representing the number of times EN node j rejects a task offload request;

5-2-5) offload decision duration: offload decision duration in observation period

For the edge calculation cluster i, the sum of the unloading decision duration of each task is as follows:

wherein the content of the first and second substances,

representing the set of computation tasks accepted by the SN node i,

indicating the time length of the unloading decision of the task r on the SN node i.

5-3) aiming at the comprehensive performance evaluation of the MECO-CTE model, the real task execution success rate and the resource loss rate are used as indexes, the user SoE and the unloading comprehensive utility of the system are verified, and the evaluation indexes are as follows:

5-3-1) real tasksThe execution success rate is as follows: within observation period, the success rate of real task execution

Calculating the ratio of the number of real and successfully executed tasks to the total number of tasks accepted by the cluster i for the edge:

wherein the content of the first and second substances,

and the actual task number successfully executed by the EN node j in the edge computing cluster i is shown.

5-3-2) resource loss rate: resource consumption rate in observation period

Calculating the ratio of the resources consumed by the execution of the dummy task and the sum of the resources provided by the EN node in the edge calculation cluster i:

wherein the content of the first and second substances,

representing a set of dummy tasks executed by the edge compute cluster i,

and

representing the computational and storage resources consumed by the execution of dummy tasks in the edge compute cluster i.

Analysis of Experimental Effect

1) Analyzing an unloading request behavior evaluation model: comparing and analyzing the TCO cooperative service system and the AODV random unloading cooperative service system, wherein the change trend of the effective utilization rate of the resources is shown in figure 8; a task request of the AODV random unloading cooperative service system adopts a random receiving strategy, and the effective utilization rate of resources fluctuates up and down on a horizontal line; the TCO cooperative service system carries out false filtering on the task request, the effective utilization rate of resources is gradually improved and finally tends to a stable state, the average effective utilization rate of the resources is 78.60%, and the average effective utilization rate of the resources is improved by 9.48%.

In the TCO collaborative service system, the impact of different offload request trust thresholds on non-cooperative node identification, false task filtering and network resource utilization in edge computing is shown in fig. 9-11; with the increase of the trust threshold of the SN node, the non-cooperative node in the edge calculation is gradually identified, and the identification rate of the non-cooperative node, the rejection rate of the unloading request and the effective utilization rate of resources are improved; when the trust threshold is less than 0.4, the accuracy rate of non-cooperative node identification and task unloading request rejection is 100%; when the trust degree is more than 0.4, the cooperative node with lower trust degree can not meet the trust requirement, and is judged as a non-cooperative node by mistake, and the accuracy begins to be reduced; when the unloading request trust threshold is 0.5, the task filtering effect is optimal, the recognition rate of the non-cooperative nodes reaches 28%, the recognition accuracy is 95.83% at the lowest, the unloading request rejection rate reaches 32.84%, the rejection accuracy is 88.36% at the lowest, and the maximum effective utilization rate of resources reaches 81.52%.

In the TCO collaborative service system, different network scale environment settings are shown in table 2, and the effects on non-cooperative node identification, false task filtering, and network resource utilization in edge computing are shown in fig. 12-14; the more the number of edge computing clusters in the edge computing is, the higher the recognition rate of non-cooperative nodes and the rejection rate of unloading requests are; when the number of the edge computing clusters is less than 70, the identification accuracy of the non-cooperative nodes and the rejection accuracy of the task unloading requests are both 100%, and when the number of the edge computing clusters is more than 70, the edge computing environment is increasingly complex along with the increase of the network scale, and the accuracy and the effective resource utilization rate are reduced.

2) Analyzing an unloading cooperation comprehensive evaluation model: comparing and analyzing the CNCO collaborative service system with a Random Walk, AODV Random unloading collaborative service system and a TOO collaborative service system, and comparing the online time, the unloading rate, the failure rate, the task rejection times and the decision time of the edge computing cluster as shown in FIGS. 15-18 and Table 3; from the graph it is found that:

a. due to the unloading randomness of the Random Walk, AODV Random unloading cooperative service system and the unloading optimization of the TOO cooperative service system, the phenomena of EN node overload and task multiple unloading failure exist in the edge computing cluster, and the online time and the task unloading rate are low; compared with other systems, the CNCO collaborative service system effectively prolongs 147.49%, 146.65% and 156.76% of online time and reduces 33.84%, 33.84% and 29.26% of task execution failure rate at the cost of decision time and task off-cluster unloading;

b. the time overhead generated by the unloading negotiation among the nodes can be measured by the task rejection times, the CNCO cooperative service system considers the task rejection behavior of the nodes, reduces the average rejection times of the tasks to 831 times, and compared with other systems, the average rejection times of the tasks are respectively reduced by 35.48%, 21.53% and 52.92%, and the time overhead generated by the multiple unloading decision of the tasks is effectively reduced.

In a CNCO (collaborative service) system, a designed PS-SDP algorithm is compared and analyzed with a Particle Swarm Optimization (PSO) algorithm, a cuckoo search algorithm and a Particle Swarm algorithm based on a simulated annealing Metropolis criterion (a solution space exists in the form of a Particle Swarm, an optimal collaboration node is searched by the PSO algorithm, and a current solution of a Particle is determined by the simulated annealing Metropolis criterion), and data pairs such as data pairs shown in FIGS. 19-21 and Table 4 are provided for excitation and reliable node behavior trust, task execution failure rate change trend and decision duration; compared with other three optimization algorithms, the PS-SDP algorithm has faster and higher convergence speed and trust degree on behavior trust change of the excitation node and the reliable node, and the average trust degree reaches 0.766 and 0.693 respectively; from time complexity analysis, the PS-SDP algorithm accepts new solutions according to the search and discovery probability, increases certain time complexity, but is only second to the PSO algorithm; from the analysis of the task execution failure rate, the average execution failure rate of the PS-SDP algorithm reaches 23.98 percent, and compared with other algorithms, the PS-SDP algorithm respectively reduces 0.99 percent, 0.62 percent and 0.27 percent, and the task unloading service quality is better; from the analysis, for the calculation task with higher unloading reliability requirement, the unloading service of the cooperative node obtained by the PS-SDP algorithm is more reliable.

3) Unloading comprehensive utility analysis: according to the success rate of executing the real tasks of the sub-network and the resource loss rate, the MECO-CTE cooperative service system and the AODV random unloading, TOO, TCO, TOO-TC and CNCO cooperative service system are compared and analyzed, the success rate of executing the real tasks, the change trend of the resource loss rate and the data comparison are respectively shown in FIGS. 22-23 and Table 5, and the following results are found from the chart:

a. based on AODV random unloading and pairwise comparison of TCO, TOO-TC, CNCO and MECO-CTE collaborative service systems, the effectiveness of a task filtering mechanism based on unloading request behavior evaluation on system comprehensive performance optimization is verified, the success rate of executing real tasks is respectively improved by 5.38%, 5.95% and 9.72%, and the resource loss rate is reduced by 3.26%, 5.13% and 3.59%.

b. Based on AODV random unloading and CNCO, TCO and MECO-CTE cooperation, the effectiveness of unloading service decisions based on unloading cooperation comprehensive evaluation on system comprehensive performance optimization is verified, the success rate of executing real tasks is respectively improved by 20.64% and 24.98%, and the resource loss rate is reduced by 1.64% and 1.97%.

The real task execution success rate of the MECO-CTE collaborative service system is 59.22% at most, and is 55.52% on average, which is 30.36% higher than that of the AODV-based random unloading collaborative service system and 27.55% higher than that of the TOO collaborative service system; the lowest resource loss rate is 6.35 percent, the average resource loss rate reaches 7.53 percent, and the resource loss rate is 5.23 percent lower than that of a random unloading cooperative service system based on AODV and 7.93 percent lower than that of a TOO cooperative service system.

By the task unloading method based on the comprehensive trust evaluation, the real task execution success rate of the edge computing cluster is effectively improved, the resource loss rate is reduced, and the performance optimization of the offshore edge computing task unloading cooperative service is realized.

To summarize: the resource margin and the task execution capacity of the node have important influence on the service quality of the offshore edge computing unloading cooperative service; the system construction unloading cooperative service has bad node identification capability, false tasks of non-cooperative nodes in the network can be effectively filtered, the best cooperative node is selected to execute the unloading service, and delay overhead and resource loss caused by the false tasks and bad cooperative behaviors in edge calculation are inhibited; by constructing the task unloading model based on the comprehensive trust evaluation, the problem of credibility of a task source is solved, the unloading cooperative service quality of the edge node is guaranteed, the purposes of reliable tasks and reliable nodes are achieved, the safe cooperation, load balance and effective utilization of resources of offshore edge calculation are realized, and the unloading comprehensive utility is improved.

TABLE 1 offloading collaborative services system configuration parameters

Table 2 different network scale environment settings

TABLE 3 offload strategy Effect comparison

TABLE 4 comparison of decision duration data under different cooperative node optimization algorithms

TABLE 5 data comparison of different task offload cooperative service systems

Claims (3)

1. A maritime edge computing unloading method based on comprehensive trust evaluation is characterized by comprising the steps of establishing a node trust forgetting function and a node behavior reward and punishment operator, introducing a task unloading request and node unloading cooperative evaluation mechanism, and constructing a task unloading cooperative service system to realize false task filtering and reliable unloading service of a maritime edge computing network;

the task unloading cooperative service system takes a land data center as a cloud server, a base station and a floating platform as super nodes SN, and an offshore vessel and a terminal intelligent device as edge nodes EN; locally aggregating EN nodes by taking the SN as a center to construct an edge computing cluster, realizing edge computing resource sharing and providing task unloading cooperative service;

in the task unloading cooperative service system, an SN node constructs an inter-node unloading request and a cooperative trust evaluation model according to task authenticity and node unloading cooperative behavior characteristics as constraints, and identifies and filters false tasks and bad cooperative nodes in a network;

in the task unloading cooperative service system, an SN node unloads cooperative trust according to cooperative candidate nodes, a selected cooperative node constructs a cooperative candidate node set, and an improved particle swarm algorithm based on search discovery probability is designed to solve the problem of maximum unloaded cooperative trust, so that load balance and efficient unloading of the cooperative nodes are realized;

the method specifically comprises the following steps:

1) a source node initiates a task unloading request to an SN node;

2) the SN node constructs an unloading request behavior evaluation model according to task authenticity, calculates unloading request trust of a source node, judges node cooperation and filters false tasks of non-cooperative nodes;

3) the SN node evaluates the resource trust and the unloading behavior trust of the EN node according to the parameters of the EN node, such as the calculation storage capacity, the task execution success rate, the execution efficiency and the like, and constructs an unloading cooperation comprehensive evaluation model; setting a trust threshold value, and selecting a trusted EN node to construct an unloading cooperation candidate node set; designing a PS-SDP algorithm to solve the optimal unloading cooperation EN node by taking the unloading cooperation trust maximization as a target;

4) the SN node sends an unloading request to the EN node; the EN node autonomously selects to accept or refuse to respond to the task unloading request from the SN node, and if the node accepts the request, the EN node executes and completes the task unloading service; otherwise, the SN node marks the EN node, and searches the cooperative node again by using the PS-SDP algorithm;

5) the EN node returns the execution result of the task and feeds back the real characteristic information of the task to the SN node; updating node behavior and task characteristic information;

6) updating the SN node unloading request trust level according to the task real characteristic information fed back by the EN node; recalculating the node behavior trust according to the unloading behavior characteristics of the current EN node; and judging whether the behavior trust of the node is lower than a neutral value, if so, generating a group of test task sets to be executed by the node, implementing excitation, and calculating and updating the behavior trust of the node again.

2. The offshore edge computing offloading method of comprehensive trust evaluation according to claim 1, wherein in step 2), the SN node constructs an offloading request behavior evaluation model according to task authenticity, calculates offloading request trust of a source node, discriminates node cooperation, and filters false tasks of non-cooperative nodes, specifically comprising the steps of:

2-1) setting initial trust degree and trust threshold of SN node i unloading request trust

2-2) constructing an unloading request behavior evaluation model, and judging an unloading request trust relationship between the unloading request behavior evaluation model and a request node:

task unloading is a dynamic decision process with time relevance, and the trust characteristics of the edge computing nodes have time attenuation attributes; constructing a decay function f (t) of the trusted memory retention of the edge computing node with respect to time t based on Ebbinghaus human brain memory forgetting theory:

dividing each observation period into t_kConstructing a historical trust forgetting function of the edge computing node in each time slice as follows:

wherein T(s) represents the trust of the s-th observation period, Δ s represents the observation period interval number, σ represents a forgetting factor, and λ represents the retention rate of the historical trust;

let the number of real tasks and dummy tasks unloaded by node j in edge computing cluster i (edge computing cluster with SN node i as center) during s observation period be

Calculating the task real rate of the SN node j according to the task execution condition in the cluster

Comprises the following steps:

in order to inhibit adverse behaviors of non-cooperative node manufacturing, unloading false tasks and the like, improve trust confidence coefficient of cooperative nodes and construct reward and punishment operators of node unloading request behaviors

Introducing a history trust forgetting function, and calculating the unloading request trust of the node j after s observation periods

Wherein the content of the first and second substances,

a reward and penalty operator representing the behavior of the offload request,

representing the actual task trust of the node i to the node j during the s observation period;

2-3) based on offload request trust threshold

Judging the cooperative relationship between the SN node i and the task unloading source node j; if it is

The node j is a non-cooperative node, the node i rejects the task unloading request of the node j, and 2-4 is switched to), otherwise, the task unloading service is executed;

2-4) starting an unloading service decision mechanism, selecting an optimal cooperative node, executing a task by the cooperative node and judging the authenticity of the task, and updating real characteristic information of the task by the SN node according to a feedback result of the cooperative node;

and 2-5) if the observation period is over, updating the unloading request trust of each SN node according to the real characteristic information of the task and by combining the historical trust relationship.

3. The offshore edge computing unloading method for comprehensive trust evaluation according to claim 1, wherein in step 3), the SN node evaluates the resource trust and the unloading behavior trust thereof, constructs an unloading cooperation comprehensive evaluation model, selects a trusted EN node to construct an unloading cooperation candidate node set, designs a PS-SDP algorithm to solve the optimal unloading cooperation EN node, and comprises the following steps:

3-1) setting initial trust degree and trust threshold of SN node i for unloading cooperative trust

3-2) updating the resource trust T according to the residual resources of the EN node_i ^resFederated behavioral trust

Compute offload collaboration trust

The specific modeling process for offloading the collaboration trust is as follows:

3-2-1) compute node resource trust T based on EN node residual compute and store resources_i ^res：

The maximum load rate of the node is set as rho, and the given node i can provide computing resources and storage resources which are respectively C_i，M_iWhen unloading task r, node i resource trust degree T_i ^resComprises the following steps:

wherein, c_iAnd m_iRepresenting the current remaining computing and storage resources of node i, c_rAnd m_rRepresents the computational and memory resources required for task r execution;

in order to avoid the influence of congestion, jitter and the like on the network stability, the unloading and the execution of tasks are completed in the same observation period, and the online time of the node i is set as

After task r is executed, its remaining online time

Comprises the following steps:

the EN node completes the computational task while remaining online during the observation period, with offloading limited to:

wherein, t_cycThe length of the observation period is indicated,

which represents the computing power of the node i,

representing the elapsed time before the task is executed during the observation period;

3-2-2) SN node takes task execution efficiency and success rate of node as constraints to construct node behavior trust model

Task execution efficiency of EN node j in edge computing cluster

The calculation formula is as follows:

the number of tasks received by the node j in the cluster in the s observation period is known as

The number of tasks accepted and successfully executed is

Task execution success rate

The calculation formula is as follows:

in order to inhibit the adverse behaviors of EN nodes in the cluster, stimulate the node integrity service, and set the reward and punishment operator of the node unloading cooperative behavior

Calculating task execution trust T of node i to node j in s observation period^suc(s)：

Wherein, ω represents the weight occupied by the task execution success rate,

representing the number of tasks unloaded from the SN node i to the node j in the s observation period;

combining node task execution efficiency

And execution trust T^suc(s) calculating the behavioral trust of the node j in the s observation period

Introducing a history trust forgetting function, and calculating the behavior trust of the node j after s observation periods

3-2-3) incorporate node resource trust T_i ^resAnd behavioral trust

Building an offload collaboration trust model

3-3) obtaining EN node unloading cooperative trust by the formula (15)

The node trust relationship is judged according to the SN node unloading cooperative trust threshold value, if so, the node trust relationship is judged

The EN node j cooperative service is credible, and the unloading cooperative candidate node set is obtained by screening

3-4) converting the optimization problem of the unloading cooperative nodes into a target optimization problem of the unloading cooperative trust maximization, and collecting nodes

Designing a PS-SDP algorithm for solving a solution space, and constructing an unloading cooperative node optimizing search target and constraint conditions of a task r in an SN node i:

performing optimization search on cooperative nodes by using a particle swarm optimization algorithm, wherein the flight speed v of a particle l in a search space_lAnd position

The update formula is as follows:

wherein v is_l(τ) and x_l(τ) represents the velocity and candidate nodes of the particle l after τ iterations,

and x^gb(τ) represents the optimal trust nodes for particle l and the entire population of particles after τ iterations, c₁And c₂Learning factor representing the particles themselves and populations, w represents an inertia factor, w_maxAnd w_minRespectively representing maximum and minimum thresholds of the inertia factors, and G is the maximum iteration number of the algorithm; introducing search finding probability p E [0,1 ∈ ]]When a number from 0 to 1 is randomly generated and is smaller than the discovery probability, discarding the current solution and searching a new solution again;

in step 3-4), the PS-SDP algorithm solves the problem of the maximum uninstalling collaboration trust, and the specific steps are as follows:

3-4-1) setting the maximum rejection times of the tasks;

3-4-2) judging whether the number of times of task unloading is maximum, if so, directly unloading the task to other edge metersCalculating clusters; otherwise, initializing node optimization parameters including particle number, iteration number tau, maximum iteration number and optimization speed v_l(τ) maximum and minimum values, learning factor c₁And c₂Maximum value of inertia factor w_maxAnd a minimum value w_minEach particle represents a potential offload collaboration candidate node; in offloading collaborative candidate node sets

Computing particle candidate node offload cooperative trust according to equation (15)

Obtaining an initial particle swarm;

3-4-3) judging whether the tau reaches the maximum iteration times; if the maximum iteration times are not reached, updating the particle flight speed and the candidate nodes according to the formula (17) to generate a new particle swarm; carrying out unloading cooperation trust evaluation on the candidate node of each particle of the new particle swarm, and searching a new optimal unloading cooperation trust node according to the formula (16); judging whether the probability p is found in accordance with the search, if so, randomly updating the particle candidate nodes to generate a new particle swarm, performing unloading cooperative trust evaluation on each particle candidate node, and updating the optimal unloading cooperative trust node; otherwise, entering the next iteration, wherein tau is tau + 1;

3-4-4) stopping iterative computation when the iteration times tau reach the maximum value to obtain the best trust node of the group; taking the node as a cooperative node and initiating an unloading request;

3-4-5) judging whether the cooperative node refuses the unloading request, if so, turning to 3-4-2), and searching a new cooperative node again; otherwise, performing task unloading service.

Images