Resource allocation method for multi-device cooperative task of industrial digital twin edge network

Abstract

The invention discloses an industrial digital twin edge network multi-device cooperative task resource allocation method, which comprises the following steps of: constructing a double-layer digital twin edge network task unloading framework which comprises a physical entity layer and a digital twin layer; establishing a comprehensive system comprising a multi-device collaborative task model, a sensing model, a communication and calculation model and a time delay and energy consumption model in the physical entity layer; based on the comprehensive system, an SCC-DT model is established for a sensor, an edge server and an actuator on the digital twinborn layer on the basis of sensing resource deviation, computing resource deviation and communication resource deviation; and based on the double-layer digital twin edge network task unloading framework, jointly optimizing calculation frequency, transmitting power, transmission bandwidth and unloading factors so as to minimize end-to-end time delay of a multi-device cooperative task. According to the method, the problem that most existing task unloading and resource allocation schemes are difficult to directly apply is solved, and the end-to-end time delay of the MDC task is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communications and industrial Internet of Things, and in particular relates to a resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network. Background Art

[0002] With the rapid development of wireless communications and the Industrial Internet of Things (IIoT), the demand for applications such as driverless cars and industrial automation has increased dramatically. These applications have huge demands for computing resources and strict requirements on latency. In order to meet these challenges, Mobile Edge Computing (MEC) technology has been introduced into IIoT. By providing cloud computing capabilities at the edge of the network, resources are migrated to a location closer to the user. MEC uses task offloading technology to offload tasks to the edge. In addition, the concept of Digital Twin Edge Network (DITEN), which combines Digital Twin (DT) and Edge Network (EN), has also been proposed. The DT model is established and maintained by collecting real-time information through edge nodes, and task offloading and resource allocation are optimized in the digital domain.

[0003] Although existing technologies have made some progress in IIoT through MEC and DITEN, there are still some shortcomings, including the following:

[0004] 1. The existing DITEN architecture only considers the computing resource bias when building the DT model, ignoring the equally important communication and perception resource bias.

[0005] 2. Existing research mainly focuses on how to offload tasks from terminal devices to edge servers for execution, without fully considering the needs of multi-device data collaboration in IIoT scenarios. Multi-device collaborative tasks (MDC) are different from traditional local independent tasks in terms of formation and execution logic, making current task offloading and resource allocation solutions difficult to directly apply.

[0006] 3. Most existing works only consider the case where the task initiating device and the result receiving device are the same device, ignoring the scenario where tasks and results are transmitted between different devices, which is an important consideration in actual industrial unmanned operations. Summary of the invention

[0007] In order to solve the above technical problems, the present invention proposes a resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network. By combining MEC and DT technologies, the problems of multi-device collaborative task offloading and resource allocation in IIoT are studied to solve the limitations of the existing technology.

[0008] To achieve the above object, the present invention provides a resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network, comprising:

[0009] Build a two-layer digital twin edge network task offloading framework, including the physical entity layer and the digital twin layer;

[0010] Establishing a comprehensive system including a multi-device collaborative task model, a perception model, a communication and computing model, and a delay and energy consumption model at the physical entity layer;

[0011] Based on the comprehensive system, SCC-DT models are established for sensors, edge servers, and actuators at the digital twin layer based on perception resource deviation, computing resource deviation, and communication resource deviation;

[0012] Based on the two-layer digital twin edge network task offloading framework, the computing frequency, transmission power, transmission bandwidth and offloading factor are jointly optimized to minimize the end-to-end delay of multi-device collaborative tasks.

[0013] Preferably, the physical entity layer includes: a perception layer, a MEC control center layer and an execution layer;

[0014] The digital twin layer consists of a digital twin of the physical entity and a mirror image of the entire wireless communication environment.

[0015] Preferably, the SCC-DT model includes:

[0016] The DT model of sensor k is expressed as:

[0017] DT k ={ST k ,ΔST k};

[0018] Among them, ST k Indicates the status information of sensor k, ΔST k Indicates DT deviation, ΔST k ={Δf k ,Δp k ,Δb k ,Δs k} respectively represent the calculation frequency deviation, transmission power deviation, bandwidth deviation and perception capability deviation between sensor k and its DT;

[0019] The DT model of actuator m is expressed as:

[0020] DT m ={ST m ,ΔST m};

[0021] Among them, ST m Indicates the S state information of actuator m, ΔST m ={Δp m ,Δb m} represents the transmission power deviation and bandwidth deviation between the actuator and its DT;

[0022] The DT model of the MEC control center layer is expressed as:

[0023] DT ME3C ={ST ME3C ,DS ME3C};

[0024] Among them, ST ME3C Indicates the real-time status data of ME3C, DS ME3C Indicates system information stored in ME3C, which is the MEC control center layer.

[0025] Preferably, the multi-device collaborative task model includes data perception, raw data unloading, data preprocessing, preprocessed data unloading, task synthesis, edge computing and result downloading.

[0026] Preferably, the multi-device collaborative task is expressed as:

[0027] J MT =(D MT ,K MT ,M MT ,C MT ,T max );

[0028] Among them, D MT , K MT and M MT They represent the data item set, sensor set, and actuator set required for the task, respectively. MT and T max Indicates the number of CPU cycles required for ME3C to calculate unit data and the maximum tolerable latency of the task.

[0029] Preferably, the perception model includes:

[0030] The amount of sensor k’s perception data is:

[0031]

[0032] Among them, s kIndicates the ability to perceive, To perceive time;

[0033] The actual amount of sensor k’s perception data is:

[0034]

[0035] Among them, Δs k is the perception bias of sensor k;

[0036] The perceived energy consumption is:

[0037] Preferably, the joint optimization problem includes: delay constraint, energy consumption constraint, bandwidth constraint, transmission power constraint, local computing frequency constraint, and offloading factor constraint.

[0038] Preferably, the process of solving the optimization problem is:

[0039] Decouple the original problem into four sub-problems;

[0040] Based on the solution of four sub-problems, the ICA-LD-AO algorithm is obtained by combining the alternating optimization algorithm of the interior convex approximation method and the Lagrangian dual method;

[0041] Based on the ICA-LD-AO algorithm, the problem of multi-device collaborative task computing offloading and resource allocation is solved.

[0042] Preferably, the alternating optimization algorithm includes local resource optimization, data transmission optimization, offloading decision optimization, and download process optimization.

[0043] The present invention also provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network as described above is implemented.

[0044] Compared with the prior art, the present invention has the following advantages and technical effects:

[0045] The present invention provides a resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network, comprising: first, constructing a two-layer digital twin edge network task offloading framework, including a physical entity layer and a digital twin layer; secondly, establishing a comprehensive system including a multi-device collaborative task model, a perception model, a communication and computing model, and a delay and energy consumption model in the physical entity layer; then, based on the comprehensive system, establishing an SCC-DT model for sensors, edge servers, and actuators in the digital twin layer based on perception resource deviation, computing resource deviation, and communication resource deviation; finally, jointly optimizing computing frequency, transmission power, transmission bandwidth, and offloading factor to minimize the end-to-end delay of multi-device collaborative tasks.

[0046] The present invention constructs a two-layer digital twin edge network task offloading framework including a physical layer and a digital layer. The method can grasp the operating status of industrial equipment more comprehensively. The two-layer structure enables resource allocation and task offloading strategies to be optimized directly in the digital domain, reducing the waste of physical resources and avoiding frequent communication between edge servers and industrial equipment.

[0047] The present invention takes into account the needs of multi-device data collaboration in IIoT scenarios, such as cross-camera tracking, multi-workshop intelligent logistics distribution, etc. Such tasks require multiple sensors to collect data and the edge server to aggregate multiple data channels, which is different from the traditional local independent tasks in terms of formation and execution logic. The present invention is specially designed for such multi-device collaborative tasks (MDC), solving the problem that most existing task offloading and resource allocation solutions are difficult to directly apply.

[0048] The present invention effectively reduces the end-to-end delay of the MDC task by jointly optimizing the calculation frequency, transmission power, transmission bandwidth and unloading factor. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0050] Figure 1 It is a DITEN system model in the IIoT scenario of an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a multi-device collaborative task execution process according to an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of system delay constraint analysis according to an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of the relationship between transmission bandwidth and end-to-end delay in an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of the relationship between the local calculation frequency and the end-to-end delay according to an embodiment of the present invention;

[0055] Figure 6 Schematic diagram of the impact of different data deviation rates on end-to-end delay in an embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram of the impact of different numbers of sensors on end-to-end delay according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0058] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0059] Embodiment 1

[0060] This embodiment provides a resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network, including:

[0061] S1. Build a two-layer digital twin edge network task offloading framework, including the physical entity layer and the digital twin layer;

[0062] Specifically, Figure 1 ,This embodiment builds a digital twin edge network (DITEN) system in the IIoT scenario. The DITEN system includes a physical entity layer and a digital twin layer.

[0063] S2, establishing a comprehensive system including a multi-device collaborative task model, a perception model, a communication and computing model, and a delay and energy consumption model at the physical entity layer;

[0064] Specifically, the physical entity layer consists of the perception layer, the MEC Control Center (MEC Control Center, ME3C) layer and the execution layer. The perception layer consists of a large number of sensors, which are responsible for sensing and collecting environmental data and have local preprocessing functions. The ME3C layer consists of a base station (BS) equipped with a MEC server, which is responsible for receiving the unloaded data from the sensor, integrating it into computing tasks and processing it. The execution layer consists of multiple actuators, which are responsible for receiving the calculation results and performing corresponding operations.

[0065] S3. Based on the comprehensive system, establish SCC-DT models for sensors, edge servers, and actuators at the digital twin layer based on perception resource deviation, computing resource deviation, and communication resource deviation;

[0066] Specifically, S31, SCC-DT model;

[0067] The DT layer consists of a digital twin of the physical entity and a mirror of the entire wireless communication environment. It is responsible for monitoring the operating status of the physical system and optimizing task offloading and resource allocation through real-time interaction. and express.

[0068] The DT model is constructed and maintained at the ME3C layer, and optimizes resource configuration by monitoring system status and simulating network operation. Due to factors such as the time-varying nature and interference of the channel, data deviations will inevitably occur between the DT model and its physical entity, and data cannot be kept completely synchronized. These deviations exist not only in computing resources, but also in communication and perception resources. Therefore, the SCC-DT model established in this embodiment comprehensively considers these three resource deviations. The DT model of sensor k is expressed as:

[0069] DT k ={ST k ,ΔST k} (1)

[0070] Among them, ST k Indicates the status information of sensor k (calculation frequency, power, etc.), ΔST k Indicates DT deviation, ΔST k ={Δf k ,Δp k ,Δb k ,Δs k} respectively represent the calculation frequency deviation, transmission power deviation, bandwidth deviation and perception capability deviation between sensor k and its DT.

[0071] The DT model of actuator m is expressed as:

[0072] DT m ={ST m ,ΔST m} (2)

[0073] Among them, ST m represents the S-state information (bandwidth, channel status, etc.) of actuator m, ΔST m ={Δp m ,Δb m} represents the transmission power deviation and bandwidth deviation between the actuator and its DT.

[0074] The DT model of ME3C is expressed as:

[0075] DT ME3C ={ST ME3C ,DS ME3C} (3)

[0076] Among them, ST ME3C Indicates the real-time status data of ME3C (CPU calculation frequency, task generation status), DS ME3C Represents system information stored by ME3C, such as the sensing range of sensors, the location of actuators, and the amount of task data.

[0077] S32, multi-device collaborative task model;

[0078] This embodiment considers tasks such as cross-camera trajectory tracking in IIoT scenarios. This type of task is different from traditional time-driven tasks. Time-driven tasks will generate and require a task to be completed in each fixed time slot. The MDC task of this embodiment is an event-driven task. The task is generated by a specific event. For example, the task is only generated when the person to be tracked enters the monitoring area, and it does not have to follow a fixed time slot. Therefore, compared with the time-driven solution, the event-driven task has a lower implementation complexity, especially when the task completion time is much longer than the time slot duration. Figure 2 As shown in the figure, the event-driven MDC task requires the collaboration of sensors, ME3C and actuators. The task execution process consists of seven stages: data perception, raw data unloading, data preprocessing, preprocessed data unloading, task synthesis, edge computing and result downloading.

[0079] The MDC task initiated by ME3C is represented by J MT =(D MT ,K MT ,M MT ,C MT ,T max ), where D MT , K MT and M MT Respectively represent the data item set, sensor set, and actuator set required for the task, satisfying in Represents the set of all data items, N k and N m represents the number of sensors and actuators associated with the task. MT and T max Indicates the number of CPU cycles required for ME3C to calculate unit data and the maximum tolerable latency of the task.

[0080] Since each sensor has a different location and perception capability, the data it perceives is also different. Assume that the set of all data that can be perceived in the system is The data that sensor k can perceive is In addition, it is assumed that the sensors are reasonably distributed and there will not be a situation where different sensors perceive the same data.

[0081] S33, perception model;

[0082] The sensor’s sensing capability set is S = {s1,…,s K}, where s k Indicates the ability to perceive, is the perception time, then the amount of perception data of sensor k is:

[0083]

[0084] From the DT model, we know that the perception deviation of sensor k is Δs k , then the actual amount of sensor k’s perception data is:

[0085]

[0086] Therefore, the perceived energy consumption is:

[0087]

[0088] S34, communication model;

[0089] (1) Uplink transmission. It represents the bandwidth allocated when sensor k communicates with ME3C. represents the channel gain between sensor k and ME3C, where β0 represents the path loss at the reference distance d0, represents the actual distance from sensor k to ME3C, and ε represents the path loss coefficient. This embodiment adopts the FDMA protocol to avoid co-channel interference. Therefore, the uplink transmission rate is:

[0090]

[0091] Among them, N0 and p k denote the noise power spectral density and the transmission power of sensor k respectively.

[0092] α k is the unloading factor of sensor k, that is Data needs to be pre-processed locally. The data is directly unloaded to ME3C. Then the DT estimated delay of the raw data unloaded by sensor k is:

[0093]

[0094] The bandwidth deviation Δb can be obtained from the SCC-DT model: k and transmit power deviation Δp k Therefore, the delay deviation of sensor k unloading raw data is:

[0095]

[0096] In the formula It means that the actual signal-to-interference-noise ratio is taken into account. Therefore, the actual delay of sensor k unloading raw data is:

[0097]

[0098] Assume βk is the ratio of the data volume after sensor k preprocessing to that before preprocessing, then the DT estimated delay of sensor k unloading preprocessed data is:

[0099]

[0100] Similar to equation (9), the delay deviation of sensor k unloading preprocessed data to ME3C is:

[0101]

[0102] Therefore, the actual delay of sensor k to offload preprocessed data is:

[0103]

[0104] The energy consumption of the data unloading process can be calculated from the delay of raw data unloading and preprocessed data unloading:

[0105]

[0106] (2) Downlink transmission. ME3C receives data from multiple sensors and synthesizes the computing tasks and completes the calculations, and finally sends the calculation results to some actuators. The bandwidth allocated to actuator m is b m ,satisfy Similar to the above, represents the channel gain between ME3C and actuator m. m represents the power allocated to actuator m, satisfying Then the downlink transmission rate from ME3C to actuator m is:

[0107]

[0108] Where N0 represents the noise power spectral density.

[0109] Assuming that the ratio of the calculation result to the task amount is γ, the calculation result can be obtained as in represents the amount of data sent from sensor k to ME3C. Therefore, the download delay from ME3C to actuator m is:

[0110]

[0111] The bandwidth deviation Δp of the downlink process can be obtained from the SCC-DT model: m and power deviation Δb m Therefore, the download delay deviation of actuator m is:

[0112]

[0113] In the formula represents the signal-to-interference-to-noise ratio of the downlink process. Therefore, the delay of executor m downloading the ME3C calculation result is:

[0114]

[0115] S35, computational model;

[0116] (1) Local preprocessing. The sensor performs preprocessing operations such as denoising and filtering on some data locally. The amount of preprocessed data of sensor k is f k represents the local processing rate, C k represents the number of CPU cycles required to preprocess a unit bit of data. Therefore, the local preprocessing delay and energy consumption of sensor k are:

[0117]

[0118] Where Δf k represents the calculation frequency deviation in the DT model, ξ k Represents constants related to the device hardware structure.

[0119] (2) Edge computing. After receiving the offloaded data from the sensor, ME3C first integrates the data to form a computing task with a task volume of The synthesis time is very short and can be ignored. Therefore, the delay of ME3C processing task is:

[0120]

[0121] S36, delay and energy consumption model;

[0122] (1) Delay model. Since pre-processed data unloading and raw data unloading cannot be performed simultaneously, there will be Figure 3 Two situations are shown. Figure 3 Case 1 in (a) indicates that the original data offloading delay is greater than the preprocessing delay, that is, At this time, the preprocessed data needs to wait for a while before it can be transmitted. Figure 3 Case 2 in (b) indicates that the original data unloading delay is less than the preprocessing delay, that is, At this time, the channel will be idle and wait for the preprocessing to complete before transmission. Combining the two situations, the delay of sensor k unloading data to ME3C is

[0123] The multi-device collaborative task considered in this embodiment cannot discard the unloaded data of any sensor, because the absence of any data will cause the task synthesis to fail. Therefore, the data unloading delay is determined by the unloading delay of the slowest sensor. Similarly, the download delay is determined by the download delay of the slowest executor, so the total delay is jointly determined by the maximum upload delay and the maximum download delay in 3(b). Therefore, the total end-to-end delay is expressed as:

[0124]

[0125] (2) Energy consumption model. Since ME3C has sufficient energy supply, the energy consumption of edge computing and result downloading can be ignored. Therefore, the total energy consumption is composed of perception energy consumption, preprocessing energy consumption, and data unloading energy consumption, which can be expressed as:

[0126]

[0127] S4. Jointly optimize the calculation frequency, transmission power, transmission bandwidth, and offloading factor to minimize the end-to-end delay of multi-device collaborative tasks.

[0128] Specifically include: S41, optimization problem proposed;

[0129] This embodiment considers seven stages of MDC task from initiation to result download. Due to the complex delay structure and the existence of delay conflicts as shown in Fig.3, the goal of this embodiment is to optimize the offloading factor α = {α k}, local calculation frequency f = {f k} and transmit power Uplink bandwidth The transmission power allocated by ME3C is p={p m} and downlink bandwidth b = {b m} to minimize the E2E delay of the MDC task. The optimization problem P1 is expressed as:

[0130]

[0131] In problem P1, (24b) and (24c) represent the delay and energy consumption constraints respectively; (24d) constrains the bandwidth value; (24e) represents the power constraint; (24f) constrains the value of the local computing frequency; (24g) constrains the value of the offloading factor; (24h) represents some non-zero constraints. Since problem P1 is a non-convex problem, it is difficult to solve directly. Therefore, this embodiment first decouples the original problem into four independent sub-problems, and then uses the internal convex approximation method (ICA) and the Lagrangian dual method (LD) to convert the non-convex problem into a convex problem, and then adopts the alternating optimization (AO) method to iteratively solve it.

[0132] S42, based on ICA-HA-AO algorithm to solve the problem of computing offloading and resource allocation of multi-device collaborative tasks;

[0133] In problem P1, the perception and edge computing delays do not involve optimization variables and can be considered as known quantities. Therefore, we use and They represent the delay of the uplink process and the download process respectively, and the E2E delay can be rewritten as:

[0134]

[0135] Optimizing the uplink transmission delay actually means ensuring that all sensors can up Complete data transmission within Similarly, it is also necessary to ensure that all actuators are within the time length τ down The result reception is completed within 10 seconds, which satisfies Therefore, let τ = {τ up ,τ down}, then the original problem P1 can be rewritten as problem P2:

[0136]

[0137] Problem P2 is still non-convex. Let's solve it by decomposing the problem.

[0138] S4201, local resource optimization;

[0139] This subproblem is solved by fixing To find the next optimal local computing and communication resource It can be expressed as SP1:

[0140]

[0141] Constraints (26c) and (24c) in SP1 are non-convex constraints. The ICA method is used to transform SP1 into a convex problem. Constraint (26c) can be split into the following two constraints:

[0142]

[0143] In practice, due to the limited local computing power, the proportion of the preprocessing part is usually very small, that is, β is much smaller than 1. Therefore, it can be reasonably ignored. Then (28) and (29) are simplified as:

[0144]

[0145] Formula (30) is still non-convex, so we introduce the variable satisfy Then formula (30) is equivalent to:

[0146]

[0147] In this case (32) is a convex constraint. Substituting into (24b), the constraint can be equivalent to:

[0148]

[0149] At this time, constraint (33a) is still non-convex. From the ICA method, we know that when x>0, y>0, the upper bound of f(x,y)=xy is convex at the point It can be given by the following inequality:

[0150]

[0151] According to (34), let x = p k , (33a) can be iteratively expressed as the following convex constraint:

[0152]

[0153] Based on the above transformation, the non-convex constraint (33a) is transformed into constraints (33b) and (35), and the non-convex constraint (30) is transformed into (32). Therefore, the non-convex problem SP1 is transformed into the convex problem SP1-C:

[0154]

[0155] SP1-C can be effectively solved by using the interior point optimization method. Therefore, the algorithm for solving problem SP1 is summarized as Algorithm 1.

[0156]

[0157]

[0158] S4202, data transmission optimization;

[0159] In this subproblem, fixed value, and find the next optimal uplink bandwidth Stated as issue SP2:

[0160]

[0161] Problem SP2 is a convex problem and can be solved using convex optimization methods such as Newton’s method

[27] .

[0162] S4203, unloading decision optimization;

[0163] In this subproblem, fixed value, solve the next optimal unloading decision (α), expressed as problem SP3:

[0164]

[0165] Problem SP3 is a problem in which all constraints are linear constraints and can be easily solved using linear programming methods.

[0166] S4204, download process optimization;

[0167] In this subproblem, fixed To find the next optimal downlink computing communication resource ( b , p ). Stated as issue SP4:

[0168]

[0169] To facilitate calculation, firstly (b m -Δb m ) and (p m -Δp m ) is equivalent to b m and p m , and then take the DT deviation into account after obtaining the optimal value. It can be seen that For p m is a concave function, and is the function g m (·) is the perspective function, so the function It's about b m and p m Therefore, problem SP4 can be solved by Lagrange dual method. Assume that η ≥ 0 represents the dual variable of constraint (39a), γ ≥ 0 represents the dual variable of constraint (39b), μ m ≥0 represents the dual variable of the mth constraint in constraint (39c). Therefore, the Lagrangian function of problem SP4 is:

[0170]

[0171] Therefore, the Lagrangian even function for problem SP-4 is:

[0172]

[0173] Its dual problem can be defined as:

[0174]

[0175] SP4 satisfies the Slater condition, so there is a strong duality between this problem and its dual problem, so solving SP4 is equivalent to solving problem (42). Therefore, we first solve problem (41) to obtain the dual function of SP4, and then maximize the dual function to obtain the solution of problem (42). When a set of {μ m}≥0, η≥0 and γ≥0, problem (41) can be decomposed into the following N m +1 sub-question:

[0176]

[0177] For subproblem (43), it is easy to obtain the optimal solution given by the following formula:

[0178]

[0179] For problem (44), it can be easily proved that when η=γ=μ m = 0, any b m >0,p m >0 is the optimal solution to the problem; when η>0,γ=μ m = 0, b m =0,p m >0 is the optimal solution; when γ>0, η=μ m = 0, p m =0 is the optimal solution. Obviously, the above cases have no practical significance, so the following focuses on the case when γ>0, μ m When >0, find the optimal solution to problem (44).

[0180] Theorem 1: When η>0,γ>0,{μ m >0}, {b m} and {p m The optimal solution of} satisfies:

[0181]

[0182] in Symbols |·| * The meaning is

[0183] When getting and After that, we can solve problem (42). Problem (42) is convex but not necessarily differentiable, so we consider solving it by the ellipsoid method. Function f(η,γ,{μ m}) About the variable {μ m >0},

[0184] The subgradients for η>0 and γ>0 are and When the optimal solution of the dual problem is obtained η * and γ * Then, the formula (45) can be obtained This is the optimal solution to the original problem SP4. It is easy to prove that when the optimal solution is reached, the following conditions must be met:

[0185]

[0186] Combining equation (46) and equation (48), we can get the optimal solution for bandwidth:

[0187]

[0188] Then we can get the optimal solution of power:

[0189]

[0190] Therefore, the algorithm for solving subproblem SP4 is summarized as Algorithm 2.

[0191]

[0192]

[0193] S4205, ICA-LD-AO algorithm for solving MTOCCRA problem;

[0194] Based on the solution of the four sub-problems, an alternating optimization (AO) algorithm based on the interior convex approximation method (ICA) and the Lagrangian dual method (LD) is proposed, named ICA-LD-AO algorithm. Assume that the optimal solution of problem P1 is The algorithm is summarized as Algorithm 3.

[0195]

[0196] S43, simulation results;

[0197] Consider an IIoT edge computing system consisting of a control center equipped with a MEC server, 10 sensor devices, and 6 actuator devices. The MCC service radius is 100m, and the sensors and actuators are randomly distributed in a 50m×50m area. The simulation parameters are given in Table 1.

[0198] Table 1

[0199]

[0200] To demonstrate the effectiveness of this solution, it is compared with the following three benchmark solutions:

[0201] (1) Communication efficiency priority scheme: The CPU frequency is fixed and the semi-definite relaxation method is used to solve the problem, aiming to give priority to the optimization of the communication process.

[0202] (2) Computational efficiency priority scheme: Given bandwidth and power, the convex difference algorithm is used to minimize the E2E delay, giving priority to the optimization of the computing process.

[0203] (3) No preprocessing scheme: In this scheme, all data is offloaded to ME3C to study the impact of offloading strategy on latency.

[0204] Figure 4 The effect of transmission bandwidth on end-to-end latency is shown. The joint optimization scheme adopted in this embodiment can achieve lower latency compared with the other three benchmark schemes. When bandwidth resources are relatively limited, the performance improvement of this embodiment is more significant. When bandwidth resources are relatively sufficient, the performance improvement is slowed down, but the latency is still reduced by about 0.25s compared with the worst solution.

[0205] Figure 5 The relationship between end-to-end latency and local computing frequency is shown. When the local computing frequency is high, data tends to be preprocessed locally. Therefore, the preprocessing ratio in the figure increases as the local computing frequency increases. In addition, although the increase in local computing power will speed up the preprocessing, it will significantly increase energy consumption, resulting in a trade-off between preprocessing and energy consumption. Therefore, in order to balance latency and energy consumption, the amount of offloading must be increased, which also explains why the performance of all solutions decreases as the local computing frequency increases, but the solution of this embodiment is always better than the baseline solution.

[0206] Figure 6 The impact of different data deviation rates on end-to-end delay is demonstrated. As the data deviation rate increases, the delay also shows an increasing trend. It can be seen that data deviation has an important impact on system performance, so when building a DT model, it is necessary to fully consider multiple data deviations and reasonably set the deviation rate to ensure the accuracy of the model. Therefore, it is very meaningful to consider multiple resource deviations when establishing the SCC-DT model in this embodiment, which can provide reference value for further optimization of the DT model in the future.

[0207] Figure 7 The performance of the four solutions with different numbers of sensors is shown. The increase in the number of sensors causes an increase in the computing load, which leads to an increase in the latency of all solutions, but the solution of this embodiment has the best performance. The reason is that the solution of this embodiment can reasonably allocate computing and communication resources and balance the amount of data preprocessing and unloading. Therefore, the solution of this embodiment can be applied in scenarios with large data volume and equipment scale.

[0208] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network, characterized in that: The following steps are involved: Build a two-layer digital twin edge network task offloading framework, including the physical entity layer and the digital twin layer; Establishing a comprehensive system including a multi-device collaborative task model, a perception model, a communication and computing model, and a delay and energy consumption model at the physical entity layer; Based on the comprehensive system, SCC-DT models are established for sensors, edge servers, and actuators at the digital twin layer based on perception resource deviation, computing resource deviation, and communication resource deviation; Based on the two-layer digital twin edge network task offloading framework, the computing frequency, transmission power, transmission bandwidth and offloading factor are jointly optimized to minimize the end-to-end delay of multi-device collaborative tasks.

2. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The physical entity layer includes: a perception layer, a MEC control center layer and an execution layer; The digital twin layer consists of a digital twin of the physical entity and a mirror image of the entire wireless communication environment.

3. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 2 is characterized in that: The SCC-DT model includes: The DT model of sensor k is expressed as: DT k ={ST k ,ΔST k }; Among them, ST k Indicates the status information of sensor k, ΔST k Indicates DT deviation, ΔST k ={Δf k ,Δp k ,Δb k ,Δs k } respectively represent the calculation frequency deviation, transmission power deviation, bandwidth deviation and perception capability deviation between sensor k and its DT; The DT model of actuator m is expressed as: DT m ={ST m ,ΔST m }; Among them, ST m Indicates the S state information of actuator m, ΔST m ={Δp m ,Δb m } represents the transmission power deviation and bandwidth deviation between the actuator and its DT; The DT model of the MEC control center layer is expressed as: DT ME3C ={ST ME3C ,DS ME3C }; Among them, ST ME3C Indicates the real-time status data of ME3C, DS ME3C Indicates system information stored in ME3C, which is the MEC control center layer.

4. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The multi-device collaborative task model includes data perception, raw data unloading, data preprocessing, preprocessed data unloading, task synthesis, edge computing and result downloading.

5. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The multi-device collaborative task is expressed as: J MT =(D MT ,K MT ,M MT ,C MT ,T max ); Among them, D MT , K MT and M MT They represent the data item set, sensor set, and actuator set required for the task, respectively. MT and T max Indicates the number of CPU cycles required for ME3C to calculate unit data and the maximum tolerable latency of the task.

6. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The perception model includes: The amount of sensor k’s perception data is: Among them, s k Indicates the ability to perceive, To perceive time; The actual amount of sensor k’s perception data is: Among them, Δs k is the perception bias of sensor k; The perceived energy consumption is:

7. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The joint optimization problem includes: delay constraint, energy consumption constraint, bandwidth constraint, transmission power constraint, local computing frequency constraint, and offloading factor constraint.

8. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 1 is characterized in that: The solution process of the optimization problem is: Decouple the original problem into four sub-problems; Based on the solution of four sub-problems, the ICA-LD-AO algorithm is obtained by combining the alternating optimization algorithm of the interior convex approximation method and the Lagrangian dual method; Based on the ICA-LD-AO algorithm, the problem of multi-device collaborative task computing offloading and resource allocation is solved.

9. The resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network according to claim 8 is characterized in that: The alternating optimization algorithm includes local resource optimization, data transmission optimization, offloading decision optimization, and download process optimization.

10. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, which, when executed by a processor, implement the resource allocation method for multi-device collaborative tasks in an industrial digital twin edge network as described in any one of claims 1 to 9.

Images