A reliable and efficient computation offloading method for dependent tasks in industrial Internet of Things

Abstract

A reliable and efficient computing offloading method for dependent tasks in an industrial Internet of Things of the present invention comprises: step 1: constructing an IIoT system assisted by MEC, where each industrial device can execute tasks on different local cores or offload tasks to different edge servers for execution; step 2: constructing a task offloading model that minimizes latency and energy consumption for industrial equipment in an IIoT system assisted by MEC; step 3: constructing a data retransmission model, a computing model, and a scheduling model; step 4: combining the data retransmission model, the computing model, and the scheduling model, and using an improved MOEA / D algorithm to solve the computing offloading model in the industrial Internet of Things scenario. The offloading method of the present invention can dynamically adjust the task allocation strategy and quickly obtain the best offloading decision, which can significantly improve the overall performance of the IIoT system and can be widely used in MEC-assisted industrial environments for computing offloading.

Description

Technical Field

[0001] The present invention belongs to the technical field of edge computing offloading of industrial Internet of Things, and relates to a reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things. Background Art

[0002] With the rapid development of the Industrial Internet of Things (IIoT), the scale of the network continues to expand. Traditional cloud-centric networks can no longer meet the real-time, security, and reliability requirements of IIoT in the transmission and processing of massive data. Although the computing power of IIoT devices has gradually increased, their resources are limited due to their size. Therefore, in IIoT, facing computationally intensive and delay-sensitive applications, the limitation of device resources may not only lead to high latency and difficulty in meeting the quality of service (QoS) requirements, but also may cause high energy consumption problems and significantly reduce the quality of user experience (QoE). To solve this problem, edge computing (MEC) is considered to be a promising solution. The core advantage of MEC is that it pushes computing power, network control, and storage to the edge of the network (such as base stations and access points) to process computationally intensive and delay-critical applications consisting of multiple interdependent computing tasks on resource-limited IIoT devices.

[0003] The current methods for studying computational offloading in MEC-assisted IIoT mainly include methods based on optimization algorithms and methods based on machine learning. Methods based on optimization algorithms mainly use linear programming, heuristic algorithms (genetic algorithms, particle swarm algorithms, and multi-objective optimization) and game theory to solve the computational offloading problem of MEC-assisted IIoT; the characteristics of machine learning-based methods are to use the prediction and adaptive capabilities of machine learning models to dynamically adjust the offloading scheme and adapt to environmental changes, mainly including deep reinforcement learning (DRL), long short-term memory network (LSTM), etc. However, machine learning models require a large amount of data for training, the training process is very time-consuming, and it is difficult to meet the real-time requirements of the IIoT environment. The cost of real-time updating and retraining models is high.

[0004] In view of the complexity of computing offloading in IIoT networks, how to dynamically allocate tasks between industrial equipment and edge servers to minimize latency and energy consumption is a problem worth studying. At the same time, network latency and reliability are difficult to guarantee, and how to achieve stable and reliable offloading while maintaining system real-time performance is also a challenge. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art, propose a reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things, solve the problems of the existing MEC-assisted IIoT system computing offloading method ignoring joint scheduling and computing offloading and task offloading anomalies, and effectively improve the offloading performance of the IIoT system.

[0006] The present invention provides a reliable and efficient computing offloading method for dependent tasks in an industrial Internet of Things, comprising the following steps:

[0007] Step 1: Build an IIoT system assisted by MEC. Each industrial device can execute tasks on different local cores or offload them to different edge servers.

[0008] Step 2: Construct a task offloading model that minimizes latency and energy consumption for industrial equipment in the MEC-assisted IIoT system;

[0009] Step 3: Construct data retransmission model, calculation model and scheduling model;

[0010] Step 4: Combine the data retransmission model, computing model and scheduling model, and use the improved MOEA / D algorithm to solve the computing offloading model in the industrial Internet of Things scenario.

[0011] The reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things of the present invention has the following beneficial effects:

[0012] 1. The present invention formulates the joint scheduling and offloading problem of multiple interdependent computing tasks in the MEC-assisted IIoT system as a multi-objective optimization problem, aiming to simultaneously minimize the average latency (ACT) of industrial applications and the average energy consumption (AEC) of IIoT devices. The improved MOEA / D algorithm proposed in the present invention has good performance, excels in the convergence and diversity of non-dominated solutions, significantly reduces system latency and energy consumption, and can be widely used for computing offloading in MEC-assisted IIoT environments.

[0013] 2. The retransmission model constructed by the present invention takes into account the situations of packet loss and data retransmission, making the simulation environment more realistic. Two optimization schemes, re-decision and re-scheduling, are proposed for abnormal scenarios of task offloading to reduce the impact of unreliable links.

[0014] 3. The improved MOEA / D algorithm of the present invention improves the population initialization process, which can increase the diversity of the population and accelerate the convergence speed, help to jump out of the local optimum, and increase the global search capability.

[0015] 4. Update by uniform crossover and single point crossover respectively EL and EO vector, taking into account the dependencies between tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things of the present invention;

[0017] Figure 2 This is a system model diagram of the MEC-assisted IIoT system of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described in detail below with reference to the accompanying drawings.

[0019] like Figure 1 As shown, a reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things is characterized by comprising the following steps:

[0020] Step 1: Build an IIoT system assisted by MEC. Each industrial device can execute tasks on different local cores or offload them to different edge servers.

[0021] In order to realize the reliable and efficient computing offloading function of the sensing-dependent tasks in IIoT based on evolutionary multi-objective optimization, such as Figure 2 As shown in the figure, an IIoT system assisted by MEC is constructed. It includes multiple edge servers, multiple wireless access points, and multiple industrial devices. A wireless access point and multiple industrial devices form a cell. Each edge server is responsible for a cell. It is connected to multiple industrial devices through wireless access points. The wireless access points and edge servers are connected through optical fibers, and the transmission delay can be ignored. The wireless access points are connected to industrial devices through wireless.

[0022] Step 2: Construct a task offloading model that minimizes latency and energy consumption for industrial equipment in the MEC-assisted IIoT system. Specifically:

[0023] ;

[0024] ;

[0025] ;

[0026] in, ACT and AEC They represent the average latency of industrial applications and the average energy consumption of industrial equipment, respectively. Represents the total number of industrial devices in the system, u Indicates the number of cells in the system, N i Indicates i The total number of industrial equipment in the community, CT ( G i,j ) indicates the i In the community j Industrial applications on industrial equipment Gi,j The completion time, G i,j =( T i,j , D i,j ),in T i,j Indicates i In the community j A collection of tasks on industrial equipment. D i,j Indicates i In the community j The dependencies between tasks on industrial equipment; each task in the task set is modeled as a four-tuple, i In the community j The first industrial equipment k Tasks t i,j,k The four-tuple of ,in c i,j,k Indicates execution of tasks t i,j,k The number of CPU cycles required, d i,j,k and o i,j,k Respectively represent the size of input data and output data, v i,j,k Representation Task t i,j,k Deadline: N i,j Indicates i The first j Total task volume of industrial equipment; E ( t i,j,k ) represents a computing task t i,j,k Energy consumption, Representation Task t i,j,k Energy consumption executed on local industrial equipment, E ES Representation Task t i,j,k Energy consumption offloaded to edge servers; ε i,j,k Representation Task t i,j,k The binary variable of offloading takes the value of 1 when offloading to local industrial equipment and takes the value of 0 when offloading to the edge server.

[0027] The dependency constraints for task offloading are:

[0028] Constraint 1: Task execution order constraint: When a task is executed, it must be ensured that all previous tasks have been completed. On the one hand, if the task is executed on the edge server, it must be ensured that the output data obtained by the edge server processing the task has been completely transmitted back to the industrial equipment. On the other hand, if the task is executed on the industrial equipment, all previous tasks must be completed.

[0029] Constraint 2: Task execution location constraint: Tasks can be offloaded to different computing cores of local industrial equipment or to different edge servers;

[0030] Constraint 3: The actual completion time of the task must be less than or equal to the task deadline.

[0031] Step 3: Construct data retransmission model, calculation model and scheduling model.

[0032] During the specific implementation, the constructed data retransmission model is as follows:

[0033] Assume that the entire path from industrial equipment to edge server has M Link, n The transmission success rate of the link is , then the probability of transmission failure is ; When the n When data transmission fails on a link, define E c ( n ) indicates the n The estimated cost of a link transmission failure is expressed as:

[0034] ;

[0035] Expected number of transfers R t ( M ) is calculated as:

[0036] ;

[0037] in, E c ( M ) represents the total cost of all link transmission failures, Indicates the cost of successful transmission of all links; It represents the probability that all links are successfully transmitted.

[0038] During the specific implementation, the constructed calculation model is as follows:

[0039] Any industrial equipment FD are defined as a set of four tuples. i In the communityj Industrial equipment FD i,j The set of four tuples is represented as ,in Indicates industrial equipment FD i,j On the h Maximum execution frequency on the computing cores; Indicates industrial equipment FD i,j In the h Maximum energy consumption on a computing core; Indicates that when industrial equipment FD i,j Maximum energy consumption when transmitting data; Indicates that when industrial equipment FD i,j The maximum energy consumption when receiving data; assuming that C The calculation frequency scaling factor is expressed as ; Assume that each industrial device processor is equipped with h Heterogeneous computing cores, h The actual computing frequency of a computing core is defined as:

[0040] ;

[0041] No. h The actual power consumption of each computing core is:

[0042] ;

[0043] make In industrial equipment FD i,j No. h Maximum computing frequency on the computing cores The minimum completion time of the task t i,j,k In industrial equipment FD i,j No. h The actual execution time of a computing core is expressed as:

[0044] ;

[0045] Task t i,j,k In local industrial equipment FD i,j No. h The actual energy consumption of a computing core is expressed as:

[0046] ;

[0047] When the task is offloaded to the edge server, i In the community j The transmission rate of industrial equipment R i,j It is expressed as:

[0048] ;

[0049] in, B i,j Indicates i In the community j The channel bandwidth of industrial equipment Indicates i In the community j Industrial equipment FD i,j and i Wireless access points in each cell AP i The channel gain between σ 2 is the noise power.

[0050] When the task is offloaded to the edge server, the offload task t i,j,k The transmission delay is expressed as:

[0051] ;

[0052] Assume that each edge server has the same computing power, denoted as F; it can be seen that assuming that the computing resources of the edge server are infinite or equally shared, therefore, for each industrial device FD ,The allocated computing resources are determined by the offloading decision and the task size.

[0053] No. l Edge Servers ES l Assigned to FD i,j The computing resources are defined as:

[0054] ;

[0055] in, Indicates that from industrial equipment FD i,j Uninstall to ES l The total data size on which the task is offloaded and , J is the total number of edge servers, each edge server has the same computing power, denoted as F .

[0056] On edge servers ES l The time required to execute the task is expressed as:

[0057] ;

[0058] Industrial Equipment FD i,j From the edge server ES l The delay in receiving data is expressed as:

[0059] ;

[0060] Therefore, the task t i,j,k On edge servers ES l The total latency of the above execution is expressed as:

[0061] ;

[0062] Task t i,j,k The energy consumption offloaded to the edge server is expressed as:

[0063] ;

[0064] in, Indicates an offload task t i,j,k The transmission energy consumption, It is represented as the energy consumption of receiving data from the edge server. If the task t i,j,k It will be executed on the computing core of the local industrial equipment, and its preparation execution time is similar to the predecessor task set pre ( t i,j,k ), depending on the maximum completion time of the predecessor task set.

[0065] In industrial equipment FD i,j The waiting time before execution is expressed as:

[0066] ;

[0067] in, Indicates i In the community j Industrial equipment l Tasks t i,j,l Execution time on the local industrial equipment computing core, Representation Task t i,j,l The data is returned from the edge server to the FD i,j completion time.

[0068] In industrial equipment FD i,j Waiting time before uploading It is expressed as:

[0069] ;

[0070] If the task t i,j,k Offloaded to the edge server for execution, Depends on the set of predecessor tasks pre ( t i,j,k ), Representation Task The time when the data transfer to the edge server was completed.

[0071] Task t i,j,k Preparation execution time on edge servers It is expressed as:

[0072] ;

[0073] in, For the task The execution completion time on the edge server, For the task t i,j,k Preparation execution time on edge servers.

[0074] exist FD i,j Industrial Applications G i,j The final completion time is defined as CT ( G i,j ), expressed as:

[0075] ;

[0076] in, is a binary variable. When the last task of the industrial equipment When executed locally, ,otherwise ; Representation Task Time to complete execution at local industrial equipment, Representation Task The time when data reception is completed.

[0077] In addition, during task offloading, the following constraints must be followed:

[0078] ;

[0079] Among them, constraint C1 is the constraint of task execution order. For complex workflows, tasks t i,j,k Only in missions t i,j,l The task can be executed only after the execution is completed. The C2 constraint indicates the location where the task is offloaded. The task can be offloaded to the local core {1,…, h}, or offload to different edge servers{ h +1,…, h + J}. The C3 constraint is the task t i,j,k The preparation time for local execution must be greater than or equal to the set of predecessor tasks pre ( t i,j,k ) in which the maximum value of the local execution time and the device acceptance time; task t i,j,k The preparation time for execution on the edge server must be greater than or equal to the previous task pre ( t i,j,k ) is the maximum value of the execution time of the local and edge servers of the task. The C4 constraint indicates that the preparation time for the execution of the task on the edge server must be greater than or equal to the transmission completion time of the data transmission to the edge server. The C5 constraint indicates that the actual completion time of the computing task must be less than or equal to the deadline of the computing task.

[0080] During the specific implementation, the scheduling calculation model constructed is as follows:

[0081] Computational tasks t i,j,k Average execution time ET i,j,k :

[0082] ;

[0083] Computational tasks t i,j,k Waiting time RT i,j,k is the maximum completion time of all its predecessor tasks:

[0084] ;

[0085] in, Indicates i In the community j Industrial equipment l Tasks t i,j,l completion time;

[0086] Response rate RR i,j,k It is defined as the ratio of the response time to the execution time of a computing task, where the response time is the sum of the execution time and the waiting time:

[0087] ;

[0088] Priority scheduling is performed on parallel computing tasks that are ready at the same time, and the priority is determined by the highest response rate.

[0089] Step 4: Combining the data retransmission model, the calculation model and the scheduling model, the improved MOEA / D algorithm is used to solve the calculation offloading model in the industrial Internet of Things scenario. The specific steps of step 4 are:

[0090] Step 4.1: Given a representation of the solution to the offloading task problem, set a fitness function containing two optimization objectives, specifically:

[0091] Step 4.1.1: In order to balance the average latency of industrial applications and the average energy consumption of industrial equipment, the offloading task problem is defined as: min ( ACT , AEC ).

[0092] Step 4.1.2: For industrial applications G i,j =( T i,j , D i,j ), Indicates i In the community j A collection of tasks on industrial equipment; each task t i,j,k There is an execution location EL and execution order EO , which constitutes a representation of the solution to the task.

[0093] Step 4.1.3: Industrial Application G i,j The representation of the solutions of all tasks in constitutes a set of offloading decisions. Let , where the task t i,j,k Execution location , h yes FDi,j The number of computing cores on J is the number of edge servers; if loc i,j,k ≤h , indicating that the task is executed on the local industrial device, otherwise it is offloaded to the edge server for execution.

[0094] Step 4.1.4: Order ,express T i,j The execution order of tasks in i In the community j Industrial applications on industrial equipment G i,j The solution is expressed as ; In each cell, the solution to the unloading task problem is , , .

[0095] Step 4.1.5: Given the solution to the offloading task problem is X, the fitness function to be optimized is expressed as: .

[0096] Step 4.2: Initialize the weight vector, calculate the distance between the weight vector and other weight vectors, and select the closest one T neighbors, which constitute the neighborhood of the subproblem, specifically:

[0097] Step 4.2.1: Assume that the population size is N , the number of weight vectors is N , through the weight vector generation formula, by evenly dividing the interval [0,1], a two-dimensional weight vector is generated ( w 1, w 2), where:

[0098] ;

[0099] ;

[0100] Each weight vector defines a subproblem, i.e., an individual in the population, T To represent the number of neighbors for each subproblem, by calculating the corresponding weight vector of any two individuals The Euclidean distance between them is used to sort the neighbors of each weight vector by distance and select the one closest to the subproblem T neighbors, forming the domain of the subproblem.

[0101] Step 4.2.2: After initializing the weight vector, the multi-objective optimization problem is decomposed into multiple scalar subproblems through the Chebyshev function, and the neighborhood is updated using the Chebyshev function. If there is a better solution to the subproblem corresponding to the neighbor weight vector, the current individual is updated through crossover and mutation operations.

[0102] Step 4.3: Initialize the population and initialize the maximum number of iterations and mutation probability.

[0103] In the specific implementation, half of the individuals in the population in step 4.3 randomly initialize the execution position and execution order, and the other half of the individuals in the population dynamically initialize the execution order and execution position using a joint scheduling and unloading algorithm based on response rate, specifically:

[0104] Step 4.3.1: The initial computation task is executed locally by default.

[0105] Step 4.3.2: Traverse using breadth-first search algorithm G i,j vertices that have no predecessors and add them to the queue.

[0106] Step 4.3.3: Divide the tasks added to the queue into new task sets TS , and calculate the task set TS The response ratio of each task in RR i,j,k .

[0107] Step 4.3.4: Follow RR i,j,k Sort from high to low, task set TS The higher the response ratio, the higher the priority of the task.

[0108] Step 4.3.5: For the task with the highest priority, the offloading location of the task is dynamically determined based on the comparison result between its local completion time and its completion time on the edge server.

[0109] Step 4.3.6: Repeat steps 4.3.2 to 4.3.5 to confirm G i,j The offloading location and execution order of all tasks in .

[0110] Step 4.4: Calculate the fitness function of each individual.

[0111] Step 4.5: Set the reference point, specifically:

[0112] Step 4.5.1: Create a list for each optimization target. Fitness_1 is used to store the average delay ACT value of all individuals in the population, and fitness_2 is used to store the average energy consumption AEC value of all individuals in the population.

[0113] Step 4.5.2: Traverse the entire population, extract the values ​​of the two optimization objectives for each individual, and add them to the corresponding lists fitness_1 and fitness_2.

[0114] Step 4.5.3: Sort the values ​​of the optimization objectives in the lists fitness_1 and fitness_2, and use the minimum values ​​in the lists fitness_1 and fitness_2 as reference points respectively.

[0115] During the optimization process, the invention improves the algorithm fitness by dynamically updating the reference point. Each time a new solution is generated, the performance of the solution on each target is checked. If the solution is better than the current reference point on a certain target, the corresponding reference point is updated.

[0116] Step 4.6: Perform fast non-dominated sorting on the individuals and divide the population into multiple levels.

[0117] The purpose of non-dominated sorting is to divide the population into multiple levels (called Pareto Fronts). The solution in the first front is the non-dominated solution with the highest quality in the current population and can be directly used to construct the Pareto solution set. By comparing the dominance relationship between individuals, the population is divided into multiple non-dominated levels, thus providing an optimization priority basis for the evolution process.

[0118] Step 4.7: Iterate the update process, randomly select two neighbors to perform crossover and mutation operations to generate new offspring individuals, and repeat steps 4.4-4.6 to update the reference point and neighbors.

[0119] The iterative update process includes: selection, crossover and mutation operations. Selection, for each generation of the population, randomly select two different individuals from the individual's neighbors as parents. Crossover, divided into EL Crossover, that is, execution position crossover; EO Crossover means execution position crossover. EL Crossover adopts the uniform crossover method, and each gene position is exchanged with probability p c Determine whether to exchange. EO Crossover, adopts the single-point crossover method and considers the priority between tasks.

[0120] Step 4.8: When the maximum number of iterations is reached, the best individual in the current population is output, and the optimal unloading decision is obtained.

[0121] The above description is only a preferred embodiment of the present invention and is not intended to limit the concept of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A reliable and efficient computing offloading method for dependent tasks in industrial Internet of Things, characterized in that: The following steps are involved: Step 1: Build an IIoT system assisted by MEC. Each industrial device can execute tasks on different local cores or offload them to different edge servers. Step 2: Construct a task offloading model that minimizes latency and energy consumption for industrial equipment in the MEC-assisted IIoT system. Specifically: Among them, ACT and AEC represent the average latency of industrial applications and the average energy consumption of industrial equipment, respectively. represents the total number of industrial equipment in the system, u represents the number of cells in the system, and N i represents the total number of industrial equipment in the ith cell, CT(G i,j ) represents the industrial application G on the jth industrial equipment in the i-th cell i,j The completion time, G i,j =(T i,j , D i,j ), where T i,j represents the set of tasks on the jth industrial equipment in the i-th cell, D i,j represents the dependency between tasks on the jth industrial equipment in the i-th cell; each task in the task set is modeled as a four-tuple, and the kth task t on the jth industrial equipment in the i-th cell i,j,k is a four-tuple represented by t i，j，k =(c i，j，k , d i，j，k , o i，j，k , v i，j，k ), where c i，j，k , indicating the execution of task t i,j,k The number of CPU cycles required, d i,j,k and i,j,k Respectively represent the size of input data and output data, v i,j,k Represents task t i,j,k Deadline; N i,j represents the total task volume of the jth industrial equipment in the i-th district; E(t i,j,k ) represents the computation task t i,j,k Energy consumption, Represents task t i,j,k Energy consumption executed on local industrial equipment, E ES Represents task t i,j,k Energy consumption offloaded to edge servers; ε i,j,k Represents task t i,j,k The binary variable of offloading takes the value of 1 when offloading to local industrial equipment and takes the value of 0 when offloading to the edge server; Step 3: Construct data retransmission model, calculation model and scheduling model; Step 4: Combine the data retransmission model, calculation model and scheduling model, and use the improved MOEA / D algorithm to solve the calculation offloading model in the industrial Internet of Things scenario; The data retransmission model constructed in step 3 is specifically: Assume that there are M links in the entire path from the industrial equipment to the edge server, and the transmission success rate of the nth link is g n , then the probability of transmission failure is 1-g n ; When the data transmission on the nth link fails, define E c (n) represents the estimated cost of transmission failure of the nth link, which is expressed as: E c (n)=(1-g1)+2·g1(1-g2)+…+n·g1g2…g i-1 (1-g i ); Expected number of transmissions R t The calculation formula of (M) is: Among them, E c (M) represents the total cost of all link transmission failures, Indicates the cost of successful transmission of all links; represents the probability that all links are successfully transmitted; The calculation model constructed in step 3 is specifically: Any industrial equipment FD is defined as a set of four tuples, the jth industrial equipment FD in the i-th cell is i,j The set of four tuples is represented as in Indicates industrial equipment FD i,j The maximum execution frequency on the hth computing core; Indicates industrial equipment FD i,j Maximum energy consumption on the hth computing core; Indicates that when industrial equipment FD i,j Maximum energy consumption when transmitting data; Indicates that when industrial equipment FD i,j The maximum energy consumption when receiving data; assuming there are C computation frequency scaling factors, denoted as α1, α2, ..., α C ; Assuming that the processor of each industrial device is equipped with h heterogeneous computing cores, the actual computing frequency of the hth computing core is defined as: The actual power consumption of the hth computing core is: make Indicates in industrial equipment FD i,j The maximum computing frequency on the hth computing core The minimum completion time of task t i,j,k FD in Industrial Equipment i,j The actual execution time of the hth computing core is expressed as: Task i,j,k In local industrial equipment FD i,j The actual energy consumption of the hth computing core is expressed as: When the task is offloaded to the edge server, the uplink transmission rate of the jth industrial device in the i-th cell is R i,j It is expressed as: Among them, B i,j represents the channel bandwidth of the jth industrial equipment in the i-th cell, g i，j represents the jth industrial equipment FD in the i-th cell i,j and the wireless access point AP of the ith cell i The channel gain between 2 is the noise power; When the task is offloaded to the edge server, the offload task t i,j,k The transmission delay is expressed as: The first edge server ES l Assign to FD i,j The computing resources are defined as: in, Indicates that from industrial equipment FD i,j Unload to ES l The total data size on the task where the task is offloaded is loc i，j，k = l and l∈{h+1,...,h+J}, J is the total number of edge servers, each edge server has the same computing power, denoted as F; On edge server ES l The time required to execute the task is expressed as: Industrial Equipment FD i,j From the edge server ES l The delay in receiving data is expressed as: Therefore, task t i,j,k On edge server ES l The total latency of the above execution is expressed as: Task i,j,k The energy consumption offloaded to the edge server is expressed as: in, Represented as offloading task t i,j,k The transmission energy consumption, It is represented as the energy consumption of receiving data from the edge server. If task t i,j,k It will be executed on the computing core of the local industrial equipment, and its preparation execution time is the same as the predecessor task set pre(t i,j,k ), depending on the maximum completion time of the predecessor task set; FD in Industrial Equipment i,j Waiting time before execution It is expressed as: in, represents the lth task t of the jth industrial equipment in the i-th cell i,j,l Execution time on the local industrial equipment computing core, Represents task t i,j,l The data is returned from the edge server to the FD through the wireless transmission channel i,j completion time; FD in Industrial Equipment i,j Waiting time before uploading It is expressed as: If the task t i,j,k Offloaded to the edge server for execution, Depends on the set of predecessor tasks pre(t i,j,k ), Represents task t i，j，l ∈(pre(t i，j，k ))The transmission completion time of data transmission to the edge server; Task i,j,k Preparation execution time on edge servers It is expressed as: in, For task t i，j，l ∈(pre(t i，j，k ))Execution completion time on the edge server, For task t i,j,k Preparation execution time on edge servers; In FD i,j Industrial Applications i,j The final completion time is defined as CT(G i,j ), expressed as: Among them, ε i，j，end is a binary variable. When the last task of the industrial equipment is i，j，end When executed locally, ε i，j，end =1, otherwise ε i，j，end =0; Represents task t i，j，end Time to complete execution at local industrial equipment, Represents task t i，j，end The time when data reception is completed; The scheduling model constructed in step 3 is specifically: Computational task t i,j,k The average execution time ET i,j,k : Computational task t i,j,k The waiting time RT i,j,k is the maximum completion time of all its predecessor tasks: in, represents the lth task t of the jth industrial equipment in the i-th cell i,j,l completion time; Response rate RR i,j,k It is defined as the ratio of the response time to the execution time of a computing task, where the response time is the sum of the execution time and the waiting time: Prioritize the parallel computing tasks that are ready at the same time, and the priority is determined by the highest response rate; The step 4 is specifically as follows: Step 4.1: Give a representation of the solution to the offloading task problem and set a fitness function containing two optimization objectives; Step 4.2: Initialize the weight vector, calculate the distance between the weight vector and other weight vectors, and select the nearest T neighbors to form the neighborhood of the subproblem; Step 4.3: Initialize the population, and initialize the maximum number of iterations and mutation probability; Step 4.4: Calculate the fitness function of each individual; Step 4.5: Set the reference point; Step 4.6: Perform fast non-dominated sorting on individuals to divide the population into multiple levels; Step 4.7: Iterative process, randomly select two neighbors to perform crossover and mutation operations to generate new offspring individuals, and repeat steps 4.4-4.6 to update the reference point and neighbors; Step 4.8: When the maximum number of iterations is reached, the best individual in the current population is output, and the optimal unloading decision is obtained; The step 4.1 is specifically as follows: Step 4.1.1: In order to balance the average latency of industrial applications and the average energy consumption of industrial equipment, the offloading task problem is defined as: min(ACT, AEC); Step 4.1.2: For industrial applications G i，j =(T i，j , D i，j ), represents the set of tasks on the jth industrial equipment in the i-th cell; each task t i，j，k There is an execution location EL and execution order EO, ​​which constitute the representation of the solution of the task; Step 4.1.3: Industrial Application G i，j The representation of the solutions of all tasks in constitutes a set of offloading decisions. Let The task t i，j，k Execution location loc i，j，k ∈{1,…,h,h+1,…h+J}, h is FD i，j The number of computing cores on the network, J is the number of edge servers; if loc i，j，k ≤h, indicating that the task is executed on the local industrial device, otherwise it is offloaded to the edge server for execution; Step 4.1.4: Order Indicates T i，j The execution order of tasks in the i-th cell, the industrial application G on the j-th industrial equipment i，j The solution is denoted as x(G i，j )=(EL i，j , E.O. i，j ); In each cell, the solution to the unloading task problem is: Step 4.1.5: Given the solution of the offloading task problem is X, the fitness function to be optimized is expressed as: F(X) = (f ACT (X), f AEC (X)).

2. The reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things according to claim 1 is characterized in that: The step 4.2 is specifically as follows: Step 4.2.1: Assuming the population size is N and the number of weight vectors is N, generate a two-dimensional weight vector (w1, w2) by evenly dividing the interval [0, 1] using the weight vector generation formula, where: w2=1-w1; Each weight vector defines a subproblem, i.e., an individual in the population. T is used to represent the number of neighbors of each subproblem. By calculating the corresponding weight vector δ of any two individuals m , m = 1, ..., N, sort the neighbors of each weight vector by distance, and select the T neighbors closest to the subproblem to form the domain of the subproblem; Step 4.2.2: After initializing the weight vector, the multi-objective optimization problem is decomposed into multiple scalar subproblems through the Chebyshev function, and the neighborhood is updated using the Chebyshev function. If there is a better solution to the subproblem corresponding to the neighbor weight vector, the current individual is updated through crossover and mutation operations.

3. The reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things according to claim 1 is characterized in that: The step 4.3 is specifically as follows: half of the individuals in the population randomly initialize the execution position and execution order, and the other half of the individuals in the population dynamically initialize the execution order and execution position using a joint scheduling and unloading algorithm based on response rate, specifically as follows: Step 4.3.1: The initial computing task is executed locally by default; Step 4.3.2: Traverse G using a breadth-first search algorithm i,j Find the vertices that have no predecessors and add them to the queue; Step 4.3.3: Divide the tasks added to the queue into a new task set TS and calculate the response ratio RR of each task in the task set TS i,j,k ; Step 4.3.4: Follow RR i,j,k Sort from high to low, the task with a higher response ratio in the task set TS has a higher priority; Step 4.3.5: For the task with the highest priority, dynamically determine the offloading location of the task based on the comparison result between its local completion time and its completion time on the edge server; Step 4.3.6: Repeat steps 4.3.2 to 4.3.5 to determine G i,j The offloading location and execution order of all tasks in .

4. The reliable and efficient computing offloading method for dependent tasks in the industrial Internet of Things according to claim 1 is characterized in that: The step 4.5 is specifically as follows: Step 4.5.1: Create a list for each optimization target, fitness_1 is used to store the average delay ACT value of all individuals in the population, and fitness_2 is used to store the average energy consumption AEC value of all individuals in the population; Step 4.5.2: Traverse the entire population, extract the values ​​of the two optimization objectives of each individual, and add them to the corresponding lists fitness_1 and fitness_2; Step 4.5.3: Sort the values ​​of the optimization objectives in the lists fitness_1 and fitness_2, and use the minimum values ​​in the lists fitness_1 and fitness_2 as reference points respectively.

Images