Traceable method for task replication and resource allocation in massive machine type communication networks

Abstract

The application discloses a traceable method for task replication and resource allocation in a large-scale machine type communication network, belongs to the technical field of wireless communication and mobile edge computing, and comprises the following steps: based on Lyapunov optimization theory, physical and virtual queue quantitative constraints are established, and long-term optimization is converted into a per-time-slot "drift plus penalty" minimization problem; then, the problem is decomposed into a task replication sub-problem, a resource block allocation sub-problem and an edge server queue management sub-problem, and the sub-problems are solved in sequence; an optimal task replication scheme is determined and saved as a first decision credential; a resource block allocation scheme is determined and saved as a second decision credential; an optimal edge server queue management scheme, including an edge server task processing rate and a task discard amount, is determined and saved as a third decision credential; and a queue state is updated, the decision credentials and the queue state are encapsulated as a periodic transaction and written into a blockchain to form an operation audit log; and the application realizes optimal balance of long-term operation cost and makes the decision process traceable.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and mobile edge computing technology, and specifically relates to a traceable method for task replication and resource allocation in large-scale machine-type communication networks. Background Technology

[0002] With the development of 5G and future wireless communication technologies, massive machine-type communication (mMTC) has become a core scenario supporting Internet of Things (IoT) applications. In applications such as intelligent transportation, intelligent manufacturing, and smart cities, hundreds of millions of sensors, cameras, and other devices (i.e., "task IoT devices") need to collect and process data in real time. To meet the stringent requirements of these applications for low latency and high reliability, Mobile Edge Computing (MEC) has emerged. It deploys computing resources at the network edge, allowing task IoT devices to offload computing tasks locally.

[0003] However, in scenarios involving massive machine-to-machine communication, the concurrent access of numerous devices and centralized task offloading can easily lead to edge server congestion and resource shortages. This congestion can cause severe queuing delays, packet loss, and even task dropping, significantly impacting service reliability. To improve the service reliability of latency-sensitive applications, the research field has proposed a "task replication" mechanism. This mechanism allows the same computational task to be sent simultaneously to a mobile edge computing server and multiple surrounding Service IoT devices; the task is considered complete once any node successfully completes the computation. This approach, utilizing redundant resources, can effectively reduce task latency and improve service reliability. However, it comes at the cost of consuming more network and computing resources. Without proper constraints and scheduling, unrestricted task replication can exacerbate network congestion, leading to a surge in system operating costs. Therefore, finding the optimal balance between task replication, resource allocation, and edge server queue management to minimize operating costs while ensuring service quality is a core issue in the field.

[0004] Existing technologies have limitations in handling high-concurrency loads in massive machine-type communication networks. On the one hand, while task replication can improve reliability, unconstrained replication consumes significant computational and wireless resources, potentially exacerbating resource shortages. On the other hand, current research often optimizes task replication, resource allocation, and edge server queue management independently, neglecting their dynamic coupling, making it difficult to achieve global optimization under high loads.

[0005] In existing technologies, Lipei Yang et al., in their paper "Reliability-Aware Task Replication for Mobile Edge Computing," proposed a reliability-aware task replication algorithm to address the stringent requirements of vehicular network services for ultra-low latency and high reliability. This approach recognizes that fluctuations in wireless communication links and the computational load on edge servers are the main obstacles affecting service reliability. Its technical path is as follows:

[0006] 1. Real-world network latency measurement and analysis: This solution involved field measurements of MEC server access latency in real LTE and 5G cellular networks. Analysis revealed a multi-peak distribution pattern in the access latency, and the key factor causing this phenomenon was identified as the "scheduling request cycle" in wireless communication.

[0007] 2. Service Reliability Assessment Modeling: Based on the above findings, this scheme proposes an algorithm called "equidistant segmentation clustering" to cluster the collected latency samples, thereby more accurately fitting the latency probability density function. This model utilizes real-time collected latency data and MEC server status information to accurately assess the current service reliability of each base station.

[0008] 3. Reliability-Aware Task Replication Decision: The optimization problem of this scheme is defined as minimizing the number of task replications while meeting service reliability requirements. To achieve this goal, the algorithm adopts a greedy strategy, selecting base stations one by one starting with the most reliable ones, until the overall service success rate of the combined system meets a preset threshold, thereby avoiding unnecessary resource redundancy.

[0009] While this approach offers valuable insights, it still suffers from the following major shortcomings when applied to large-scale, high-concurrency machine-type communication networks:

[0010] (1) Fragmented optimization decisions and lack of scalability: Existing studies usually treat "task replication" and "task transmission" as two separate problems, and existing models are designed for networks of limited scale, making it difficult to extend to scenarios with massive device communication. In addition, in pursuing latency reduction, these studies often ignore the fact that excessive replication will consume a large number of resource blocks, which may offset the advantages. This fragmented and small-scale optimization approach makes it impossible to achieve the global optimum.

[0011] (2) Lack of traceability and auditability of the decision-making process: Whether it is task replication or resource allocation, the existing technical solutions are generally a "black box" decision-making process. In complex collaborative scenarios involving multiple entities and platforms, once service quality disputes or resource abuse occur, the lack of a transparent and tamper-proof decision record makes it extremely difficult to hold people accountable and audit them afterward. Summary of the Invention

[0012] To overcome the shortcomings of the existing technologies, the present invention aims to provide a traceable method for task replication and resource allocation in large-scale machine-type communication networks. By uniquely combining Lyapunov optimization, mean-field game theory, and blockchain smart contract technology, the method incorporates task replication, resource block allocation, and edge server queue management into a unified joint optimization framework. The key decision-making process of resource scheduling is stored on-chain in an immutable manner, which achieves an optimal balance between service reliability and task latency while ensuring long-term operating costs and making the decision-making process traceable.

[0013] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0014] A traceable method for task replication and resource allocation in large-scale machine-type communication networks includes the following steps:

[0015] S1: Establishing a system optimization model and stability framework: Based on Lyapunov optimization theory, establishing a physical task queue for edge servers. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the long-term stochastic optimization problem is transformed into a "drift plus penalty" minimization problem for each time slot.

[0016] S2: Decompose the optimization problem within the decision-making cycle: Based on the inherent logical dependencies of the decisions, decompose the "drift plus penalty" minimization problem obtained in step S1 into a task replication subproblem, a resource block allocation subproblem, and an edge server queue management subproblem, and solve them sequentially; based on the task replication subproblem, determine the optimal task replication scheme and save it as the first decision credential; based on the resource block allocation subproblem, determine the resource block allocation scheme and save it as the second decision credential; based on the edge server queue management subproblem, determine the optimal edge server queue management scheme, including the edge server task processing rate and the task drop rate, and save it as the third decision credential.

[0017] S3: Update system state and complete periodic transaction on-chain: Based on the task replication scheme, resource block allocation scheme, and edge server queue management scheme determined in step S2, automatically invoke the smart contract deployed on the distributed ledger to update the edge server physical task queue established in step S1. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The updated state of the first decision certificate, the second decision certificate and the third decision certificate generated in step S2 is encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and enters the next decision cycle.

[0018] Step S1 specifically includes:

[0019] Step S1.1: The system optimization model minimizes the cumulative total operating cost of a single task IoT device i performing task offloading. The cumulative total operating cost consists of task transmission cost and computation cost. The optimization problem is to minimize the cumulative total operating cost under constraints such as resource block constraints, task completion constraints, server queue stability constraints, and service reliability constraints, which is expressed as optimization problem P1:

[0020]

[0021] in, For task IoT devices exist Operating costs at any given time;

[0022] Resource block constraint: The total number of resource blocks allocated to the IoT devices for transmission to the base station. And the total number of resource blocks transmitted from task IoT devices to service IoT devices. Each resource pool must not exceed its own maximum limit. and ,Right now:

[0023]

[0024] Task completion constraints: For tasks transmitted to the edge server, the following conditions must be met to ensure that the task is completed within a preset delay T:

[0025]

[0026] in, The transmission rate between task IoT device i and base station b. Let i be the transmission rate between task IoT device i and service IoT device j. To improve the speed of task processing for IoT devices, The total number of tasks generated by task IoT device i. This is the size of the result data packet returned by base station b or serving IoT device j to the task IoT device after the calculation task is completed. Let be the bit error rate (BER) between task IoT device i and service IoT device j. Maximum allowable processing delay;

[0027] Server queue stability constraint: arrival rate of the task queues on the edge server The sum cannot exceed the edge server's task processing rate for an extended period. With discard rate The sum of these values ​​ensures that the average speed of the queue remains stable, i.e., satisfies:

[0028]

[0029] Service reliability constraints: For each terminal node Its service reliability Not lower than a minimum threshold That is, satisfying Among them, service reliability Transmission error rate of each transmission link This is determined in conjunction with the server's task drop rate. It is determined by the following formula:

[0030]

[0031] Step S1.2: Using Lyapunov optimization theory, the following three types of queues are introduced to quantify the task completion constraints, server queue stability constraints, and service reliability constraints in Step S1.1:

[0032] Q(t): Physical task queue on the edge server;

[0033] H i (t): used to represent the virtual reliability queue of task IoT devices;

[0034] Z(t): Uses an ε-persistent service queue to limit the worst-case virtual queuing latency of the server queue Q;

[0035] Step S1.3: Construct the "drift plus penalty" expression, transforming the optimization problem P1 obtained in step S1.1 into a per-slot "drift plus penalty" minimization problem. The transformation steps are as follows:

[0036] First, we define a Lyapunov function L(Θ(t)) to quantify the overall instability of the system at time t. The Lyapunov function L(Θ(t)) is defined as the sum of all queues Q(t), Z(t), and H in the system. i (t) The form of the sum of squares of lengths:

[0037]

[0038] in, It includes Q(t), Z(t), and H. i (t) is a vector of all queue states. A single-slot Lyapunov drift Δ(Θ(t)) is introduced to evaluate the dynamic trend of system state changes. The single-slot Lyapunov drift Δ(Θ(t)) is defined as the expected change of the Lyapunov function from the current time t to the next time t+1, that is:

[0039]

[0040] Construct a "drift plus penalty" expression Let C be the objective function, and minimize the system's operating cost. i (t);

[0041] Here, V is a preset, non-negative control parameter used to balance the importance of cost optimization and queue stability.

[0042] The solution process in step S2 includes:

[0043] S2.1: Solve the task replication subproblem: Determine the optimal task replication scheme based on the state of the reliability virtual queue and the network channel conditions;

[0044] S2.2: Solving the resource block allocation subproblem: Model the resource block allocation subproblem as a mean-field game model to obtain the Nash equilibrium solution of the resource block allocation scheme;

[0045] S2.3: Solving the edge server queue management subproblem: The edge server dynamically decides the optimal edge server queue management scheme based on the task arrival rate and the real-time backlog of its own queue, and determines the edge server task processing rate and task discard amount.

[0046] In step S2.1, the objective function of the task replication subproblem aims to minimize a weighted sum that includes transmission and computation costs, as well as penalties arising from the backlog of the reliability virtual queue. The objective function is as follows:

[0047]

[0048] Where V is the Lyapunov control parameter. H represents the transmission and computation costs of tasks between task IoT device i and service IoT device j, respectively. i (t) represents the current backlog in the reliability virtual queue, Ω i , j (t) represents the transmission error rate;

[0049] Based on the objective function, for each task IoT device i, a service IoT device j that minimizes the objective function is selected for task replication until the maximum replication limit is reached, thus obtaining the optimal task replication target. The optimal task replication target is then used as the first decision credential and uploaded to the blockchain to ensure the traceability of task distribution behavior.

[0050] In step S2.2, mean-field game theory is used to transform the resource block allocation subproblem from an "N-body" game problem into a "single-body" optimal control problem. The Nash equilibrium solution of the "single-body" optimal control problem is obtained by iteratively solving a set of coupled partial differential equations, including:

[0051] Solve the Hamilton-Jacobi-Bellman equations to obtain the optimal resource allocation strategy for each node;

[0052] Solve the Fokker-Planck-Kolmogorov equations to obtain the system state probability distribution;

[0053] By iteratively solving the Hamilton-Jacobi-Bellman equation and the Fock-Planck-Kolmogorov equation until convergence, a stable Nash equilibrium point is found, and the resource block allocation scheme is obtained.

[0054] A dynamic adjustment mechanism is adopted for adjustment: if the channel state information of a certain transmission link is detected to be higher than the threshold, the resource block allocation scheme is directly output; if the channel state information of a certain transmission link is detected to be lower than the threshold, the system starts a preset timer and temporarily reduces the resource block allocation of the link according to the correction coefficient during the timer period. The adjusted resource block allocation is the final resource block allocation scheme.

[0055] Finally, the final resource block allocation scheme will be determined through Nash equilibrium solution and dynamic adjustment mechanism, serving as the second decision credential, and recorded in the distributed ledger through smart contract.

[0056] In step S2.3, the edge server dynamically decides the optimal task processing rate based on the task arrival rate and the backlog in its own queue. Given the task drop rate θ(t), the decision objective is to minimize a weighted sum that includes computational cost, task drop penalty, and edge server queue backlog penalty. The objective function is as follows:

[0057]

[0058] The objective function aims to minimize computational cost, and based on the queue... , and Dynamically adjust task discard amount based on length and edge server task processing speed :

[0059] (1) Task drop rate :

[0060] By introducing the remaining amount of the task A(t), Add the current queue backlog Q(t) to the total number of new tasks arriving in this period. Subtract the planned workload for this period. The formula is as follows:

[0061]

[0062] A(t) represents the amount of tasks that will accumulate in the queue at the end of the current time slot if the server does not perform any task discarding operation.

[0063] The decision method for the task discard amount θ(t) is as follows: Based on the remaining task amount A(t), the server adopts a segmented threshold strategy to determine the task discard amount θ(t):

[0064]

[0065] The method of judgment is as follows:

[0066] If A(t) exceeds the server's maximum drop-out capacity If so, the server will drop the packets to the maximum extent possible;

[0067] If A(t) is within a controllable range, then the amount of discarded tasks is exactly equal to the remaining amount of tasks A(t), and the goal is to empty the queue when the time slot ends.

[0068] like Then there is no need to discard any tasks;

[0069] (2) Edge server task processing rate :

[0070] This decision involves a dynamic trade-off between processing costs and queue backlog pressure; the system assesses the total queue pressure comprised of both physical task queues and virtual delay queues.

[0071] Ultimately, the output edge server task processing rate The amount of task discarded, θ(t), will be recorded in the distributed ledger via a smart contract as a third decision credential.

[0072] In step S3, after all decisions for a decision cycle are made, the system updates the status of all physical and virtual queues based on the decision results of this cycle. At the same time, all decisions generated by the task IoT devices and edge servers in this cycle, together with the updated queue status, are encapsulated into a periodic transaction and written into a distributed ledger to form an operation audit log that includes decision timestamps, task IoT device or edge server identifiers, task replication schemes, resource block allocation schemes, edge server queue management schemes, and tamper-proof operation audit logs.

[0073] This invention also provides a traceable system for task replication and resource allocation in large-scale machine-type communication networks, comprising:

[0074] The system optimization model and stability framework establishment module is used to establish a physical task queue for edge servers based on Lyapunov optimization theory. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the long-term stochastic optimization problem is transformed into a "drift plus penalty" minimization problem for each time slot.

[0075] The optimization problem decomposition module within the decision-making cycle is used to decompose the "drift plus penalty" minimization problem into three sub-problems based on the inherent logical dependencies of the decision: task replication, resource block allocation, and edge server queue management. These sub-problems are solved sequentially. Based on the task replication sub-problem, the optimal task replication scheme is determined and saved as the first decision credential. Based on the resource block allocation sub-problem, the optimal resource block allocation scheme is determined and saved as the second decision credential. Based on the edge server queue management sub-problem, the optimal edge server queue management scheme, including edge server task processing rate and task drop rate, is determined and saved as the third decision credential.

[0076] The system status and periodic transaction on-chain update module is used to automatically invoke smart contracts deployed on the distributed ledger to update the physical task queues of edge servers based on task replication schemes, resource block allocation schemes, and edge server queue management schemes. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The generated first decision certificate, second decision certificate and third decision certificate are encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and then enter the next decision cycle.

[0077] This invention also provides a traceable device for task replication and resource allocation in large-scale machine-type communication networks, comprising:

[0078] Memory: A computer program that stores the traceable methods for task replication and resource allocation in the aforementioned large-scale machine-type communication network, and is a computer-readable device;

[0079] Processor: Used to implement the traceable method for task replication and resource allocation in large-scale machine-type communication network when executing the computer program.

[0080] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the traceable method for task replication and resource allocation in large-scale machine-type communication networks.

[0081] Compared with the prior art, the present invention has the following advantages:

[0082] (1) Establishing a unified joint optimization framework: This invention aims to create a joint optimization framework that integrates task replication, resource block allocation, and edge server queue management. By using Lyapunov optimization theory, these three interrelated sub-problems are placed under a unified objective for collaborative decision-making, thereby breaking down the barriers between task replication strategy and underlying transmission reliability, and achieving system-level end-to-end performance optimization.

[0083] (2) Efficient and scalable large-scale resource allocation mechanism: To address the challenge of resource block competition among massive devices, this invention innovatively introduces mean-field game theory to model the resource block allocation subproblem. This theory approximates the complex multi-agent "N-body" game problem into a computationally easier-to-handle single-agent optimal control problem, greatly reducing the solution complexity. This makes the resource allocation mechanism of this invention highly scalable and can be efficiently applied to massive machine communication networks.

[0084] (3) Proactive System Stability Assurance and Resource Consumption Balance: This invention optimizes edge server queue management as an independent core sub-problem, solving the problem of the lack of proactive management and intervention mechanisms for server queues in existing technologies. Edge servers can proactively and dynamically decide on the optimal edge server task processing rate and "task discarding strategy" based on task arrival rate and their own queue backlog, effectively preventing edge server congestion and fundamentally ensuring system stability under high concurrency loads. Simultaneously, since resource block allocation is included in the long-term total operating cost, the resource consumption caused by task replication is effectively constrained, achieving improved service reliability while avoiding excessive consumption of resource blocks.

[0085] (4) Unique full-process traceability and auditing capability for decision-making: This invention innovatively introduces smart contract technology, solving the technical gaps in existing technologies such as opaque decision-making processes and lack of recording and verification mechanisms. The method treats the task replication scheme, resource block allocation scheme, and edge server queue management scheme as immutable credentials, which are recorded in a distributed ledger through smart contracts. This constructs a complete and traceable operation audit log chain for the entire resource scheduling process, enabling precise traceability of task processing paths and resource block allocation, and greatly improving the transparency, security, and credibility of the system.

[0086] In summary, compared with existing technologies, this invention innovatively combines Lyapunov optimization, mean-field game theory, and blockchain smart contract technology to construct a unified joint optimization framework. This framework enables collaborative decision-making in task replication, resource block allocation, and edge server queue management, breaking down the barriers between upper-layer strategies and lower-layer transmission reliability. It innovatively introduces mean-field game theory, transforming complex multi-agent games into single-agent optimal control problems, significantly reducing solution complexity and enabling the resource allocation mechanism to achieve high scalability for massive numbers of devices. Through proactive edge server queue management, it effectively constrains resource consumption while ensuring system stability. Furthermore, it utilizes distributed ledger technology to achieve transparency, immutability, and traceability throughout the decision-making process, constructing a complete operation audit log chain, significantly improving the system's security and reliability. Attached Figure Description

[0087] Figure 1 This is a schematic diagram of the overall process of the method of the present invention.

[0088] Figure 2 This is a schematic diagram of the resource block allocation process of the method of the present invention.

[0089] Figure 3 This is a schematic diagram of the edge server queue management sub-process of the method of the present invention.

[0090] Figure 4 This is a schematic diagram of the traceable decision chain evidence storage process of the present invention.

[0091] Figure 5 This is a schematic diagram comparing the cumulative total operating cost over time under different strategies in an embodiment of the present invention.

[0092] Figure 6 This is a schematic diagram comparing the service reliability over time under different strategies according to an embodiment of the present invention.

[0093] Figure 7 This is a comparative diagram showing the changes in edge server task queue backlog over time under different strategies according to an embodiment of the present invention.

[0094] Figure 8 This is a schematic diagram comparing the total cumulative operating cost and system service reliability under different timer duration settings in an embodiment of the present invention.

[0095] Figure 9 This is a schematic diagram comparing the backlog of edge server task queues with the maximum cumulative transmission task queue under different timer duration settings in an embodiment of the present invention. Detailed Implementation

[0096] The technical solution adopted by the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0097] To systematically address the aforementioned challenges, this invention innovatively designs a three-in-one collaborative control architecture. This architecture consists of three core technological pillars:

[0098] (1) Control framework based on Lyapunov optimization: As the top-level decision architecture of the system, its core function is to transform a stochastic optimization problem with the goal of optimizing long-term operating costs into a series of deterministic "drift plus penalty" minimization problems that can be solved independently in each decision cycle by introducing a virtual queue to quantify service quality constraints.

[0099] (2) Resource allocation model based on mean-field game: As a core game theory tool for solving resource block competition among large-scale nodes, this model approximates the N-body game problem, which is highly coupled among massive nodes and computationally difficult to handle, into a single-body optimal control problem that is easy to solve. Each node only needs to formulate its own optimal resource allocation strategy based on the average effect of the collective behavior of all other nodes in the system, thereby greatly reducing the solution complexity and ensuring the high scalability of the solution in large-scale dynamic networks.

[0100] (3) Traceable control mechanism based on blockchain and smart contracts: As the technical foundation for ensuring the transparency and auditability of system decisions, this mechanism introduces distributed ledger technology to encapsulate the key decisions generated by the aforementioned optimization framework and game model into immutable periodic transactions for on-chain storage through smart contracts, thereby constructing a complete operation audit log chain for the entire resource scheduling process.

[0101] like Figure 1 As shown, a traceable method for task replication and resource allocation in large-scale machine-type communication networks includes the following steps:

[0102] S1: Establishing a system optimization model and stability framework: Based on Lyapunov optimization theory, establishing a physical task queue for edge servers. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the complex long-term stochastic optimization problem is transformed into a deterministic "drift plus penalty" minimization problem that can be solved in every time slot.

[0103] S2: Decompose the optimization problem within the decision-making cycle: Based on the inherent logical dependencies of the decisions, decompose the "drift plus penalty" minimization problem obtained in step S1 into a task replication subproblem, a resource block allocation subproblem, and an edge server queue management subproblem, and solve them sequentially; based on the task replication subproblem, determine the optimal task replication scheme and save it as the first decision credential; based on the resource block allocation subproblem, determine the resource block allocation scheme and save it as the second decision credential; based on the edge server queue management subproblem, determine the optimal edge server queue management scheme, including the edge server task processing rate and the task drop rate, and save it as the third decision credential.

[0104] S3: Update system state and complete periodic transaction on-chain: Based on the task replication scheme, resource block allocation scheme, and edge server queue management scheme determined in step S2, automatically invoke the smart contract deployed on the distributed ledger to update the edge server physical task queue established in step S1. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The updated state of the first decision certificate, the second decision certificate and the third decision certificate generated in step S2 is encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and enters the next decision cycle.

[0105] Step S1 specifically includes:

[0106] Step S1.1: The core objective of the system optimization model is to minimize the cumulative total operating cost of a single task IoT device i performing task offloading. The cumulative total operating cost consists of task transmission cost and computation cost. The optimization problem is to minimize the cumulative total operating cost under the constraints of resource block constraints, task completion constraints, server queue stability constraints, and service reliability constraints. Formalized, this is expressed as optimization problem P1:

[0107]

[0108] in, For task IoT devices exist Operating costs at any given time;

[0109] Resource block constraint: The total number of resource blocks allocated to the IoT devices for transmission to the base station. And the total number of resource blocks transmitted from task IoT devices to service IoT devices. Each resource pool must not exceed its own maximum limit. and ,Right now:

[0110]

[0111] Task completion constraints: For tasks transmitted to the edge server, the following conditions must be met to ensure that the task is completed within the preset delay T:

[0112]

[0113] in, The transmission rate between task IoT device i and base station b. Let i be the transmission rate between task IoT device i and service IoT device j. To improve the speed of task processing for IoT devices, The total number of tasks generated by task IoT device i. This refers to the size of the result data packet returned by the computing node (base station b) or the serving IoT device j to the task IoT device after the computation task is completed. Let be the bit error rate (BER) between task IoT device i and service IoT device j. Maximum allowable processing delay;

[0114] Server queue stability constraint: arrival rate of the task queues on the edge server The sum cannot exceed the edge server's task processing rate for an extended period. With discard rate The sum of these values ​​ensures that the average speed of the queue remains stable, i.e., satisfies:

[0115]

[0116] Service reliability constraints: For each terminal node Its service reliability It must be no less than a minimum threshold That is, satisfying Among them, service reliability Transmission error rate of each transmission link This is determined in conjunction with the server's task drop rate. It is determined by the following formula:

[0117] ;

[0118] Step S1.2: To solve the optimization problem P1 and ensure queue stability, this invention employs Lyapunov optimization theory and introduces the following three types of queues to quantify the task completion constraints, server queue stability constraints, and service reliability constraints in step S1.1:

[0119] Q(t): Physical task queue on the edge server;

[0120] H i (t): used to represent the virtual reliability queue of task IoT devices;

[0121] Z(t): Uses an ε-persistent service queue to limit the worst-case virtual queuing latency of the server queue Q;

[0122] Step S1.3: Construct the "drift plus penalty" expression, cleverly transforming the optimization problem P1 obtained in step S1.1 into a deterministic "drift plus penalty" minimization problem solvable in every time slot. The transformation steps are as follows:

[0123] First, we define a Lyapunov function L(Θ(t)) to quantify the overall instability of the system at time t. The Lyapunov function L(Θ(t)) is defined as the sum of all queues Q(t), Z(t), and H in the system. i (t) The form of the sum of squares of lengths:

[0124]

[0125] in, It includes Q(t), Z(t), and H. i (t) is a vector of all queue states. A single-slot Lyapunov drift Δ(Θ(t)) is introduced to evaluate the dynamic trend of system state changes. The single-slot Lyapunov drift Δ(Θ(t)) is defined as the expected change of the Lyapunov function from the current time t to the next time t+1, that is:

[0126]

[0127] The core idea of ​​Lyapunov optimization is not merely to pursue queue stability (i.e., minimize Δ(Θ(t))), but rather to further minimize the system's operating cost C. i (t) (i.e., the "penalty" term) For this purpose, a "drift plus penalty" expression is constructed. The objective function is...

[0128] Here, V is a preset, non-negative control parameter used to balance the importance of cost optimization and queue stability: a larger V value means that the algorithm will focus more on minimizing operating costs and may tolerate a moderate increase in queue length; a smaller V value means that the algorithm will focus more on maintaining queue stability and may sacrifice some cost optimality.

[0129] In step S2, since the "drift plus penalty" expression to be optimized in each decision cycle contains multiple highly coupled decision variables such as task replication, resource block allocation, and server management, directly solving them jointly is computationally very difficult. Therefore, this invention adopts an alternating iterative optimization method, which decomposes the complex joint optimization problem into three sub-problems with clear objectives and easier computation based on the inherent logical dependencies between the decision variables.

[0130] This decomposition follows a clear decision-making sequence: First, the destination of tasks needs to be determined, i.e., task replication (subproblem P2.1); second, after the communication links are determined, resource blocks are allocated to these links, i.e., resource block allocation (subproblem P2.2); finally, the result of resource block allocation determines the rate at which tasks reach the edge server, allowing the server to make the optimal response and dynamically manage its own queue, i.e., edge server queue management (subproblem P2.3). These three subproblems will be solved sequentially in the subsequent steps S2.1, S2.2, and S2.3.

[0131] The solution process in step S2 includes:

[0132] S2.1: Solve the task replication subproblem: Determine the optimal task replication scheme based on the state of the reliability virtual queue and the network channel conditions;

[0133] S2.2: Solving the resource block allocation subproblem: To solve the resource competition among a large number of devices, the resource block allocation subproblem is modeled as an average field game model to obtain the Nash equilibrium solution of the resource block allocation scheme;

[0134] S2.3: Solving the edge server queue management subproblem: The edge server dynamically decides the optimal edge server queue management scheme based on the task arrival rate and the real-time backlog of its own queue, and determines the edge server task processing rate and task discard amount.

[0135] In step S2.1, the objective function of the task replication subproblem aims to minimize a weighted sum that includes transmission and computation costs, as well as penalties arising from reliability virtual queue backlogs. The objective function is as follows:

[0136] P2.1:

[0137] Where V is the Lyapunov control parameter. H represents the transmission and computation costs of tasks between task IoT device i and service IoT device j, respectively. i (t) represents the current backlog in the reliability virtual queue, Ω i , j (t) represents the transmission error rate; this formula reveals the core trade-off in the decision-making process: when poor historical reliability leads to H... i When the value of (t) is large, a transmission error rate Ω is selected. i , j (t) Low-service IoT devices j will generate a significant negative term, namely "revenue", which incentivizes the algorithm to make decisions that improve service reliability;

[0138] Based on the objective function, for each task IoT device i, a service IoT device j that minimizes the objective function is selected for task replication until the maximum replication limit is reached, thus obtaining the optimal task replication target. The optimal task replication target is then used as the first decision credential and uploaded to the blockchain to ensure the traceability of task distribution behavior.

[0139] In step S2.2, the core task is to allocate resource blocks to these communication links after determining the transmission destinations of all tasks (sub-problem P2.2). This step aims to solve the problem of intense resource contention caused by massive concurrent communication between nodes, ensuring the fairness and efficiency of resource allocation while meeting the latency constraints of the tasks.

[0140] In large-scale machine communication scenarios, resource allocation decisions among massive numbers of devices are highly coupled, and direct solutions would result in a computationally extremely complex multi-agent game problem. To address this challenge, this invention innovatively employs mean-field game theory, transforming the resource block allocation subproblem from an "N-person" game problem into a "single-person" optimal control problem. The core idea is that when the number of participants in the game approaches infinity, the impact of a single participant's behavior on the entire system becomes negligible. Therefore, the complex "N-person" game problem can be approximated as a more easily solvable "single-person" optimal control problem, where each device only needs to make its optimal decision based on the average effect of the collective behavior of all other nodes, thereby significantly reducing computational complexity and making the solution highly scalable.

[0141] Specifically, the Nash equilibrium solution to the "single" optimal control problem can be obtained by iteratively solving a set of coupled partial differential equations, which consist of two core parts:

[0142] Hamilton-Jacobi-Bellman equation (HJB equation): This equation takes the perspective of "individual optimality" and uses dynamic programming to solve the optimal resource allocation strategy that each independent task IoT device i should take to minimize its total future cost under a given system mean field.

[0143] Fock-Planck-Kolmogorov equations (FPK equations): These equations, from a "macro-evolutionary" perspective, describe how the probability distribution of the entire system state evolves over time when all devices follow the optimal strategy calculated by the Hamilton-Jacobi-Bellman equations (HJB equations).

[0144] By iteratively solving the Hamilton-Jacobi-Bellman equation (HJB equation) and the Fock-Planck-Kolmogorov equation (FPK equation) until convergence, this invention can find a stable Nash equilibrium point, thereby obtaining a resource block allocation scheme.

[0145] Furthermore, to address the practical issue of potentially inaccurate or lost Channel State Information (CSI) in real-world networks, this step also includes a dynamic adjustment mechanism to improve the robustness of the scheme. The dynamic adjustment mechanism works as follows: if the CSI of a transmission link is detected to be higher than a threshold, a resource block allocation scheme is directly output; if the CSI of a transmission link is detected to be lower than the threshold, the system starts a preset timer and temporarily reduces the resource block allocation for that link based on a correction coefficient during the timer period. The adjusted resource block allocation is the final resource block allocation scheme. This conservative allocation strategy effectively avoids resource waste and transmission failures caused by incorrect channel estimation.

[0146] Finally, the final resource block allocation scheme is determined through Nash equilibrium solution and dynamic adjustment mechanism, serving as the second decision credential, and recorded in the distributed ledger via smart contract. This step ensures that every detail of resource allocation is transparent, immutable, and auditable. The flowchart of S2.2 is as follows: Figure 2 As shown.

[0147] In step S2.3, the edge server dynamically determines the optimal task processing rate based on the task arrival rate and the backlog in its own queue. Given the task drop rate θ(t), the decision objective is to minimize a weighted sum that includes computational cost, task drop penalty, and edge server queue backlog penalty. The objective function is as follows:

[0148] P2.3:

[0149] The objective function aims to minimize computational cost, and based on the queue... , and Dynamically adjust task discard amount based on length and edge server task processing speed :

[0150] (1) Task drop rate :

[0151] To proactively manage the queue length and prevent it from growing indefinitely, this invention introduces a task surplus A(t). Add the current queue backlog Q(t) to the total number of new tasks arriving in this period. Subtract the planned workload for this period. The formula is as follows:

[0152]

[0153] A(t) is not simply the expected queue length, but rather represents the amount of tasks that will accumulate in the queue at the end of the current time slot if the server does not perform any task discarding operation.

[0154] The decision method for the task discard amount θ(t) is as follows: Based on the remaining task amount A(t), the server adopts a segmented threshold strategy to decide the task discard amount θ(t):

[0155]

[0156] The purpose of this strategy is to alleviate queue pressure, and the judgment method is as follows:

[0157] If A(t) exceeds the server's maximum drop-out capacity If so, the server will drop the packets to the maximum extent possible;

[0158] If A(t) is within a controllable range, then the amount of discarded tasks is exactly equal to the remaining amount of tasks A(t), and the goal is to empty the queue when the time slot ends.

[0159] like Then there is no need to discard any tasks;

[0160] (2) Edge server task processing rate :

[0161] After determining the number of dropped tasks, the second step is to decide on the task processing rate of the edge servers. The core of this decision is to dynamically weigh the processing cost against the queue backlog pressure; the system will evaluate the total queue pressure, which is composed of the physical task queue Q(t) and the virtual latency queue Z(t), i.e., Q(t) + Z(t).

[0162] Define the number of valid tasks reached. This refers to the total number of new tasks that arrive during this period and are not discarded.

[0163]

[0164] Based on the above trade-off mechanism, the system compares the total queue pressure with the cost threshold and calculates the edge server task processing rate in two modes. :

[0165] Mode 1: Maximized Processing Mode

[0166] When the total queue pressure exceeds the cost threshold, it indicates a high risk of system backlog. The system then adopts a backlog cleanup strategy, setting the processing rate to the current total number of tasks plus the server's maximum task processing capacity. The smaller value:

[0167]

[0168] Mode 2: Optimal Cost Mode

[0169] When the total queue pressure is below the cost threshold, it indicates that the system is in a safe state; the system only processes the number of valid tasks arriving in the current cycle to avoid wasting computing power.

[0170]

[0171] Ultimately, this step outputs the edge server task processing rate. The task discard amount θ(t) will serve as a third decision credential, recorded in the distributed ledger via a smart contract, ensuring the transparency, auditability, and traceability of the server's behavior when dealing with network load. The flowchart for S2.3 is as follows: Figure 3 As shown.

[0172] In step S3, after all decisions for a decision cycle are made, the system updates the status of all physical and virtual queues based on the decision results of this cycle. Simultaneously, all decisions generated by the task IoT devices and edge servers in this cycle, along with the updated queue status, are encapsulated into a periodic transaction and written to the distributed ledger, forming a log containing decision timestamps, task IoT device or edge server identifiers, task replication schemes, resource block allocation schemes, edge server queue management schemes, and immutable operation audit logs. The evidence preservation process is as follows: Figure 4 As shown.

[0173] The key and core protection point of this invention lies in its originality in combining Lyapunov optimization, mean-field game theory, and blockchain technology to construct a complete and traceable method for scheduling and controlling large-scale resource blocks. Specific protection points are as follows:

[0174] (1) Joint Scheduling Framework Based on Lyapunov Optimization: This invention proposes a control framework that transforms a long-term stochastic optimization problem into a per-slot "drift plus penalty" minimization problem. Its core lies in not only establishing a physical task queue for temporary task storage but also innovatively introducing a "virtual reliability queue" and a "virtual delay queue," respectively used to quantify constraints on system stability and quality of service. This design enables the three originally independent processes of task replication, resource block allocation, and edge server queue management to be jointly optimized within a unified framework, achieving a dynamic balance between system cost and multiple quality of service constraints.

[0175] (2) A large-scale resource allocation method based on mean-field game theory: To solve the problem of resource competition among massive nodes, this invention models the resource block allocation subproblem as a mean-field game model. This method approximates the complex "N-body" game problem into an easily solvable "single-body" optimal control problem, and obtains the Nash equilibrium solution for resource allocation by iteratively solving the Hamilton-Jacobi-Bellman equation and the Fock-Planck-Kolmogorov equation. This method greatly reduces computational complexity and has high scalability, which is one of the core technologies that enable this invention to be applied to large-scale dynamic networks.

[0176] (3) Traceable control mechanism based on blockchain and smart contracts: A core innovation of this invention is the introduction of distributed ledger technology, which realizes transparency and traceability of the entire resource scheduling process. The method treats the three key outputs, namely the "task replication scheme," "resource block allocation scheme," and "edge server queue management scheme," as independent decision credentials and records them immutably in the distributed ledger through smart contracts. Finally, all credentials and the updated queue status of each decision cycle are encapsulated into a periodic transaction and put on the chain, forming a complete chain of operation audit logs with timestamps.

[0177] (4) Accurately traceable operation audit logs: The on-chain operation audit logs generated based on the aforementioned mechanism explicitly include decision timestamps, task IoT device or edge server identifiers, task replication schemes, resource block allocation schemes, and edge server queue management schemes. This makes it possible to accurately trace the transmission path of data between different nodes, thereby effectively verifying the compliance of resource sharing and data exchange processes, and providing solid technical support for achieving full lifecycle management of cross-platform data sharing.

[0178] Figure 1This is a schematic diagram of the overall process of the traceable task replication and resource scheduling method proposed in this invention. The diagram illustrates the three core steps from S1 to S3, which constitute a complete closed loop for each decision cycle. The process begins with establishing a system optimization model (S1), followed by problem decomposition (S2), and then sequentially solving three sub-problems: task replication (S2.1), resource allocation (S2.2), and edge server queue management (S2.3). Finally, the system state is updated and on-chain notarization is completed (S3), before proceeding to the next cycle.

[0179] Figure 2 This is a detailed schematic diagram of the resource allocation sub-process in the mean-field game of this invention. The flowchart elaborates on the internal implementation logic of step S2.2. The process begins by receiving the current system state and approximates the Nash equilibrium solution for resource allocation by iteratively solving the Hamilton-Jacobi-Bellman equation (HJB equation) and the Fokker-Planck-Kolmogorov equation (FPK equation). After obtaining the preliminary allocation scheme, the process also includes a crucial judgment step to detect the reliability of the channel state information (CSI) and, based on the detection result, decide whether to enable the dynamic adjustment mechanism, ultimately outputting a robustly enhanced resource block (RB) allocation scheme.

[0180] Figure 3 This diagram illustrates the detailed implementation logic of the edge server queue management sub-process in this invention. It details how the server dynamically decides on the task processing rate and drop strategy based on the task arrival rate and its own queue status. This process is a two-step decision-making process: First, based on the predicted potential queue backlog A(t), the optimal task drop rate θ(t) is determined through a segmented threshold logic; second, after determining the drop rate, the final task processing rate is determined based on whether the total queue pressure (determined jointly by the physical task queue Q(t) and the virtual delay queue Z(t)) exceeds a cost threshold. .

[0181] Figure 4 This diagram illustrates the unique traceable decision-making chain evidence storage process of this invention. It clearly demonstrates the core mechanism by which this invention achieves "traceability." At the end of a decision cycle, the three key decisions (task replication strategy, resource block allocation scheme, and server management strategy) generated by steps S2.1, S2.2, and S2.3 are packaged into independent "decision credentials," and together with the system queue state updated in step S3, are encapsulated into a structured periodic transaction. This transaction is written into the distributed ledger by calling the smart contract interface, thus forming a complete and tamper-proof operation audit log.

[0182] Experimental Analysis

[0183] Simulation conditions

[0184] To verify the effectiveness of the proposed traceable control method for task replication and resource scheduling in large-scale machine-type communication networks, this embodiment constructs a simulation platform based on Matlab 2022a. The simulation scenario is set within a 1 km × 1 km geographical area, with specific parameter settings as follows:

[0185] 1. Network topology: 50 task IoT devices are randomly deployed within the area, and 10 to 20 service IoT devices are distributed around each task device.

[0186] 2. Task parameters: The amount of task data generated by each task device is set to 500 Kb, the total time period T is set to 20 ms, and each task device can select up to 3 service IoT devices for task replication.

[0187] 3. Communication parameters:

[0188] The communication coverage radius is set to 200 m.

[0189] The bandwidth between the mission device and the base station (BS) is set to 10 MHz, and the bandwidth between the mission device and the service device is set to 5 MHz.

[0190] All transmission power is set to 23 dBm, and noise power spectral density is set to -104 dBm.

[0191] 4. Server parameters: The maximum computing speed of the edge server is set to 5 × 10^9 Hz, and the maximum task drop rate is set to 2.5 × 10^8 Hz.

[0192] 5. Data source: The movement trajectory of the device was generated using the HighD dataset to simulate a real dynamic environment.

[0193] Simulation content

[0194] This embodiment mainly verifies the performance of the present invention from the following two dimensions:

[0195] 1. Overall Performance Comparison Analysis: The method of this invention is compared with several mainstream strategies in the prior art to verify the advantages of this invention in reducing operating costs and ensuring service reliability. The comparison strategies include:

[0196] MEC-only offload strategy: Tasks are only sent to the edge server and are not replicated.

[0197] Average allocation strategy: Resource blocks are evenly distributed in the transmission link without dynamic optimization.

[0198] Locally optimal strategy: Select only the node with the best channel conditions for transmission.

[0199] Global optimal strategy: assumes the optimal solution under ideal conditions where all future channel information is known, and uses it as a reference for the upper limit of performance.

[0200] 2. Effectiveness Analysis of Dynamic Adjustment Mechanism: Regarding the "dynamic adjustment mechanism" proposed in step S2.2 of this invention, the impact of changing the timer duration parameter on the total system operating cost, service reliability, and edge server queue backlog is analyzed to verify the robustness of the mechanism when the channel state is unstable.

[0201] Simulation results

[0202] 1. Overall performance comparison results

[0203] (1) Comparison of cumulative total operating costs

[0204] The changes in cumulative total operating costs for different strategies within a 20ms time period, such as... Figure 5 As shown, the cumulative total operating cost of this invention remains at an extremely low level throughout the entire simulation period and is very close to the globally optimal strategy. Especially in the early stage of system operation, i.e., 0-10ms, the method of this invention almost coincides with the globally optimal strategy, proving that this invention can efficiently utilize resources and approach the theoretical optimal performance.

[0205] (2) Service reliability comparison

[0206] The service reliability changes under different strategies within a 20ms time period are as follows: Figure 6 As shown, the service reliability of this invention rapidly climbs to above 0.9 after a brief startup phase and remains consistently above 0.8 for the remainder of the simulation period. Even with a slight decrease later, it still outperforms most comparative strategies. This is thanks to the virtual reliability queue mechanism introduced in this invention, which can detect and compensate for fluctuations in service quality in real time.

[0207] (3) Comparison of task queue backlog on edge servers

[0208] Figure 7 shows the changes in edge server task queue backlog over time under different strategies. The edge server queue backlog of this invention remains at an extremely low level throughout the entire period, with minimal fluctuations. This indicates that the edge server queue management scheme of this invention can effectively prevent tasks from piling up on the edge server, thereby reducing queuing latency and ensuring the real-time performance of the system.

[0209] 2. Results of the effectiveness analysis of the dynamic adjustment mechanism

[0210] (1) Total cumulative operating costs and system service reliability

[0211] Figure 8 This section compares the system's total cumulative operating costs and service reliability under different timer duration settings. Figure 8 It is evident that the timer duration has a significant impact on system performance, and there exists an optimal value. When the timer duration is set to 5 seconds, the system's total cumulative operating cost reaches its minimum. Simultaneously, as the timer duration increases, the overall service reliability of the system remains almost unchanged. This is because the introduction of the timer mechanism allows more resource blocks to be allocated to transmission links specific to channel conditions, thus effectively ensuring overall system reliability during dynamic adjustments.

[0212] (2) Edge server task queue backlog and maximum cumulative transmission task queue

[0213] Figure 9 This demonstrates a comparison of edge server task queue backlog and maximum cumulative transmission task queue under different timer duration settings. For example... Figure 9 As shown, the service queue backlog of the edge server decreases with increasing timer duration. This is mainly because the adjustment factor reduces the total number of resource blocks allocated, thereby reducing the number of tasks arriving at the server per unit time and alleviating server congestion. This result verifies that by selecting an appropriate timer length, the queue length of the edge server can be effectively adjusted, thus better meeting the requirements of task latency constraints.

[0214] Simulation results show that:

[0215] 1. The method proposed in this invention is significantly superior to existing strategies such as average allocation and MEC offloading in terms of reducing operating costs, improving service reliability, and reducing queue backlog, and its performance is very close to the theoretically optimal strategy.

[0216] 2. The dynamic adjustment mechanism introduced in this invention exhibits good robustness. By reasonably setting the timer duration, congestion on edge servers can be effectively suppressed even under unstable channel conditions, while keeping operating costs to a minimum, thus verifying the application value of this mechanism in practical large-scale machine communication networks.

[0217] This invention also provides a traceable system for task replication and resource allocation in large-scale machine-type communication networks, comprising:

[0218] The system optimization model and stability framework establishment module is used to implement step S1, which establishes the physical task queue of the edge server based on Lyapunov optimization theory. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the long-term stochastic optimization problem is transformed into a "drift plus penalty" minimization problem for each time slot.

[0219] The optimization problem decomposition module within the decision-making cycle is used to decompose the "drift plus penalty" minimization problem obtained in step S1 into three sub-problems: task replication, resource block allocation, and edge server queue management, based on the inherent logical dependencies of the decision. These sub-problems are solved sequentially. Based on the task replication sub-problem, the optimal task replication scheme is determined and saved as the first decision credential. Based on the resource block allocation sub-problem, the optimal resource block allocation scheme is determined and saved as the second decision credential. Based on the edge server queue management sub-problem, the optimal edge server queue management scheme, including the edge server task processing rate and task drop rate, is determined and saved as the third decision credential.

[0220] The system status and periodic transaction on-chain update module is used to implement the task replication scheme, resource block allocation scheme, and edge server queue management scheme determined in step S2 in step S3. It automatically calls the smart contract deployed on the distributed ledger to update the edge server physical task queue established in step S1. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The updated state of the first decision certificate, the second decision certificate and the third decision certificate generated in step S2 is encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and enters the next decision cycle.

[0221] This invention also provides a traceable device for task replication and resource allocation in large-scale machine-type communication networks, comprising:

[0222] Memory: A computer program that stores the traceable methods for task replication and resource allocation in the aforementioned large-scale machine-type communication network, and is a computer-readable device;

[0223] Processor: Used to implement the traceable method for task replication and resource allocation in large-scale machine-type communication network when executing the computer program.

[0224] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the traceable method for task replication and resource allocation in large-scale machine-type communication networks.

Claims

1. A traceable method for task replication and resource allocation in large-scale machine-type communication networks, characterized in that, Includes the following steps: S1: Establishing a system optimization model and stability framework: Based on Lyapunov optimization theory, establishing a physical task queue for edge servers. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the long-term stochastic optimization problem is transformed into a "drift plus penalty" minimization problem for each time slot. Step S1 specifically includes: Step S1.1: The system optimization model minimizes the cumulative total operating cost of a single task IoT device i performing task offloading. The cumulative total operating cost consists of task transmission cost and computation cost. The optimization problem is to minimize the cumulative total operating cost under constraints such as resource block constraints, task completion constraints, server queue stability constraints, and service reliability constraints, which is expressed as optimization problem P1: in, For task IoT devices exist Operating costs at any given time; Resource block constraint: The total number of resource blocks allocated to the IoT devices for transmission to the base station. And the total number of resource blocks transmitted from task IoT devices to service IoT devices. Each resource pool must not exceed its own maximum limit. and ,Right now: Task completion constraints: For tasks transmitted to the edge server, the following conditions must be met to ensure that the task is completed within a preset delay T: in, The transmission rate between task IoT device i and base station b. Let i be the transmission rate between task IoT device i and service IoT device j. To improve the speed of task processing for IoT devices, The total number of tasks generated by task IoT device i. This is the size of the result data packet returned by base station b or serving IoT device j to the task IoT device after the calculation task is completed. Let BER be the bit error rate between task IoT device i and service IoT device j. Maximum allowable processing delay; Server queue stability constraint: arrival rate of the task queues on the edge server The sum cannot exceed the edge server's task processing rate for an extended period. With discard rate The sum of these values ​​ensures that the average speed of the queue remains stable, i.e., satisfies: Service reliability constraints: For each terminal node Its service reliability Not lower than a minimum threshold That is, satisfying Among them, service reliability Transmission error rate of each transmission link This is determined in conjunction with the server's task drop rate. It is determined by the following formula: ； Step S1.2: Using Lyapunov optimization theory, the following three types of queues are introduced to quantify the task completion constraints, server queue stability constraints, and service reliability constraints in Step S1.1: Q(t): Physical task queue on the edge server; H i (t): used to represent the virtual reliability queue of task IoT devices; Z(t): Uses an ε-persistent service queue to limit the worst-case virtual queuing latency of the server queue Q; Step S1.3: Construct the "drift plus penalty" expression, transforming the optimization problem P1 obtained in step S1.1 into a per-slot "drift plus penalty" minimization problem. The transformation steps are as follows: First, we define a Lyapunov function L(Θ(t)) to quantify the overall instability of the system at time t. The Lyapunov function L(Θ(t)) is defined as the sum of all queues Q(t), Z(t), and H in the system. i (t) The form of the sum of squares of lengths: in, It includes Q(t), Z(t), and H. i (t) is a vector of all queue states. A single-slot Lyapunov drift Δ(Θ(t)) is introduced to evaluate the dynamic trend of system state changes. The single-slot Lyapunov drift Δ(Θ(t)) is defined as the expected change of the Lyapunov function from the current time t to the next time t+1, that is: Construct a "drift plus penalty" expression Let C be the objective function, and minimize the system's operating cost. i (t); Here, V is a preset, non-negative control parameter used to balance the importance of cost optimization and queue stability; S2: Decompose the optimization problem within the decision-making cycle: Based on the inherent logical dependencies of the decisions, decompose the "drift plus penalty" minimization problem obtained in step S1 into a task replication subproblem, a resource block allocation subproblem, and an edge server queue management subproblem, and solve them sequentially; based on the task replication subproblem, determine the optimal task replication scheme and save it as the first decision credential; based on the resource block allocation subproblem, determine the resource block allocation scheme and save it as the second decision credential; based on the edge server queue management subproblem, determine the optimal edge server queue management scheme, including the edge server task processing rate and the task drop rate, and save it as the third decision credential. S3: Update system state and complete periodic transaction on-chain: Based on the task replication scheme, resource block allocation scheme, and edge server queue management scheme determined in step S2, automatically invoke the smart contract deployed on the distributed ledger to update the edge server physical task queue established in step S1. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The updated state of the first decision certificate, the second decision certificate and the third decision certificate generated in step S2 is encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and enters the next decision cycle.

2. The traceable method for task replication and resource allocation in large-scale machine-type communication networks according to claim 1, characterized in that, The solution process in step S2 includes: S2.1: Solve the task replication subproblem: Determine the optimal task replication scheme based on the state of the reliability virtual queue and the network channel conditions; S2.2: Solving the resource block allocation subproblem: Model the resource block allocation subproblem as a mean-field game model to obtain the Nash equilibrium solution of the resource block allocation scheme; S2.3: Solving the edge server queue management subproblem: The edge server dynamically decides the optimal edge server queue management scheme based on the task arrival rate and the real-time backlog of its own queue, and determines the edge server task processing rate and task discard amount.

3. The traceable method for task replication and resource allocation in large-scale machine-type communication networks according to claim 2, characterized in that, In step S2.1, the objective function of the task replication subproblem aims to minimize a weighted sum that includes transmission and computation costs, as well as penalties arising from the backlog of the reliability virtual queue. The objective function is as follows: Where V is the Lyapunov control parameter. H represents the transmission and computation costs of tasks between task IoT device i and service IoT device j, respectively. i (t) represents the current backlog in the reliability virtual queue, Ω i , j (t) represents the transmission error rate; Based on the objective function, for each task IoT device i, a service IoT device j that minimizes the objective function is selected for task replication until the maximum replication limit is reached, thus obtaining the optimal task replication target. The optimal task replication target is then used as the first decision credential and uploaded to the blockchain to ensure the traceability of task distribution behavior.

4. The traceable method for task replication and resource allocation in large-scale machine-type communication networks according to claim 2, characterized in that, In step S2.2, mean-field game theory is used to transform the resource block allocation subproblem from an "N-body" game problem into a "single-body" optimal control problem. The Nash equilibrium solution of the "single-body" optimal control problem is obtained by iteratively solving a set of coupled partial differential equations, including: Solve the Hamilton-Jacobi-Bellman equations to obtain the optimal resource allocation strategy for each node; Solve the Fokker-Planck-Kolmogorov equations to obtain the system state probability distribution; By iteratively solving the Hamilton-Jacobi-Bellman equation and the Fock-Planck-Kolmogorov equation until convergence, a stable Nash equilibrium point is found, and the resource block allocation scheme is obtained. A dynamic adjustment mechanism is adopted for adjustment: if the channel state information of a certain transmission link is detected to be higher than the threshold, the resource block allocation scheme is directly output; if the channel state information of a certain transmission link is detected to be lower than the threshold, the system starts a preset timer and temporarily reduces the resource block allocation of the link according to the correction coefficient during the timer period. The adjusted resource block allocation is the final resource block allocation scheme. Finally, the final resource block allocation scheme will be determined through Nash equilibrium solution and dynamic adjustment mechanism, serving as the second decision credential, and recorded in the distributed ledger through smart contract.

5. The traceable method for task replication and resource allocation in large-scale machine-type communication networks according to claim 2, characterized in that, In step S2.3, the edge server dynamically decides the optimal task processing rate based on the task arrival rate and the backlog in its own queue. Given the task drop rate θ(t), the decision objective is to minimize a weighted sum that includes computational cost, task drop penalty, and edge server queue backlog penalty. The objective function is as follows: The objective function aims to minimize computational cost, and based on the queue... , and Dynamically adjust task discard amount based on length and edge server task processing speed : (1) Task drop rate : By introducing the remaining amount of the task A(t), Add the current queue backlog Q(t) to the total number of new tasks arriving in this period. Subtract the planned workload for this period. The formula is as follows: A(t) represents the amount of tasks that will accumulate in the queue at the end of the current time slot if the server does not perform any task discarding operation. The decision method for the task discard amount θ(t) is as follows: Based on the remaining task amount A(t), the server adopts a segmented threshold strategy to determine the task discard amount θ(t): The method of judgment is as follows: If A(t) exceeds the server's maximum drop-out capacity If so, the server will drop the packets to the maximum extent possible; If A(t) is within a controllable range, then the amount of discarded tasks is exactly equal to the remaining amount of tasks A(t), and the goal is to empty the queue when the time slot ends. like Then there is no need to discard any tasks; (2) Edge server task processing rate : This decision involves a dynamic trade-off between processing costs and queue backlog pressure; the system assesses the total queue pressure comprised of both physical task queues and virtual delay queues. Ultimately, the output edge server task processing rate The amount of task discarded, θ(t), will be recorded in the distributed ledger via a smart contract as a third decision credential.

6. The traceable method for task replication and resource allocation in large-scale machine-type communication networks according to claim 1, characterized in that, In step S3, after all decisions for a decision cycle are made, the system updates the status of all physical and virtual queues based on the decision results of this cycle. At the same time, all decisions generated by the task IoT devices and edge servers in this cycle, together with the updated queue status, are encapsulated into a periodic transaction and written into a distributed ledger to form an operation audit log that includes decision timestamps, task IoT device or edge server identifiers, task replication schemes, resource block allocation schemes, edge server queue management schemes, and tamper-proof operation audit logs.

7. A traceable system for task replication and resource allocation in large-scale machine-type communication networks based on the method of claim 1, characterized in that, include: The system optimization model and stability framework establishment module is used to establish a physical task queue for edge servers based on Lyapunov optimization theory. Virtual Reliability Queues for IoT Devices and virtual delay queue By quantifying the system stability and service quality constraints through these three queues, the long-term stochastic optimization problem is transformed into a "drift plus penalty" minimization problem for each time slot. The optimization problem decomposition module within the decision-making cycle is used to decompose the "drift plus penalty" minimization problem into three sub-problems based on the inherent logical dependencies of the decision: task replication, resource block allocation, and edge server queue management. These sub-problems are solved sequentially. For the task replication sub-problem, the optimal task replication scheme is determined and saved as the first decision credential. For the resource block allocation sub-problem, the optimal resource block allocation scheme is determined and saved as the second decision credential. For the edge server queue management sub-problem, the optimal edge server queue management scheme, including edge server task processing rate and task drop rate, is determined and saved as the third decision credential. The system status and periodic transaction on-chain update module is used to automatically invoke smart contracts deployed on the distributed ledger to update the physical task queues of edge servers based on task replication schemes, resource block allocation schemes, and edge server queue management schemes. Virtual Reliability Queues for IoT Devices and virtual delay queue The state is updated, and the smart contract automatically performs state verification and transaction submission on the chain according to the predetermined conditions. The generated first decision certificate, second decision certificate and third decision certificate are encapsulated together as a periodic transaction, written into the blockchain, forming a traceable and tamper-proof operation audit log chain, and then enter the next decision cycle.

8. A traceable device for task replication and resource allocation in large-scale machine-type communication networks, characterized in that, include: Memory: A computer program storing the traceable method for task replication and resource allocation in large-scale machine-type communication networks as described in any one of claims 1-6, and is a computer-readable device; Processor: Used to implement the traceable method for task replication and resource allocation in a large-scale machine-type communication network as described in any one of claims 1-6 when executing a computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the traceable method for task replication and resource allocation in large-scale machine-type communication networks as described in any one of claims 1-6.

Images