Air edge computing resource allocation method and device oriented to execution uncertainty, equipment and storage medium

Abstract

The invention discloses an execution uncertainty-oriented air edge computing resource allocation method, device and equipment and a storage medium, and relates to the technical field of air edge computing, and the method comprises the steps: obtaining user number information, edge server task information, bit length information, time delay constraint information and energy constraint information; determining a target system resource allocation probability based on the user number information, the edge server task information, the bit length information, the time delay constraint information and the energy constraint information; and controlling a system to carry out resource allocation based on the target system resource allocation probability, and completing execution uncertainty-oriented air edge computing resource allocation. According to the method and the device, the uncertainty factors in the dynamic environment are quantified into probability constraints, calculation is performed through the optimal resource allocation strategies corresponding to different scenes, and the resource allocation probability of the target system is obtained, so that the system performs resource allocation in a self-adaptive manner, and the utilization efficiency of the system resources is maximized.

Description

Technical Field

[0001] This application relates to the field of over-the-air edge computing technology, and in particular to over-the-air edge computing resource allocation methods, apparatus, devices and storage media for execution uncertainty. Background Technology

[0002] The sixth-generation wireless network targets IoT scenarios such as industrial automation, environmental monitoring, and smart mobility, emphasizing massive connectivity, low latency, and high reliability. With the rapid growth of mobile terminals and data-intensive services, terminal-side computing power and energy budget have become bottlenecks. Over-the-air computing utilizes the superposition characteristics of multiple access channels to aggregate and perform function calculations on parallel uplink signals over the air interface. This couples the communication and computing processes at the architecture level, meaning that multiple devices transmit synchronously on the same time-frequency resources, and the receiving end directly performs approximate calculations of the objective function on the superimposed waveforms, thereby reducing uplink overhead and shortening end-to-end latency. However, in actual deployment, it is affected by execution uncertainty, that is, the number of CPU cycles required per bit varies randomly with application complexity and data characteristics, causing the computation latency to be randomly distributed. At the same time, over-the-air superposition is sensitive to multi-terminal synchronization and power shaping, and channel fluctuations and interference will further affect the probability of timely completion.

[0003] Currently, existing resource allocation practices are based on deterministic or long-term average assumptions, treating task workload and channel parameters as fixed values. Under constraints of total latency and total energy consumption, they collaboratively optimize transmit power, time-division resources, and task splitting ratios. However, in single-user scenarios, while the number of CPU cycles required per bit is modeled as a random variable and attempts are made to jointly optimize related parameters, this approach fails to effectively integrate with multi-user interference and system dynamics. It either optimizes communication and computational metrics separately or only considers randomness at the single-user level, failing to perform joint modeling under a unified probabilistic objective of timely completion. This makes it difficult to achieve global optimum under the dual constraints of latency and energy consumption. In multi-user scenarios, although methods such as deep reinforcement learning are being used for online scheduling, they lack precise probabilistic characterization and unified modeling of the inherent randomness of execution uncertainty, and the computational complexity is very high, making real-time adaptation impossible. Therefore, overcoming the uncertainty of task execution in airborne edge computing to achieve efficient resource allocation in single-user and multi-user scenarios under latency and energy constraints has become an urgent problem to be solved.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, device, and storage medium for allocating resources in over-the-air edge computing to address the uncertainty of task execution in over-the-air edge computing, thereby solving the technical problem of how to overcome the uncertainty of task execution in over-the-air edge computing and achieve efficient resource allocation in single-user and multi-user scenarios under latency and energy constraints.

[0006] To achieve the above objectives, this application proposes a method for allocating airborne edge computing resources in the face of execution uncertainty, the method comprising: Obtain user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; The target system resource allocation probability is determined based on the user quantity information, the edge server task information, the bit length information, the latency constraint information, and the energy constraint information. Resource allocation is performed based on the target system resource allocation probability control system to complete the allocation of airborne edge computing resources oriented towards execution uncertainty.

[0007] In one embodiment, the number of users is a single user; The step of determining the resource allocation probability of the target system based on the edge server task information, the bit length information, the latency constraint information, and the energy constraint information includes: Acquire transmission feature information, edge feature information, and local feature information; The transmission success probability, edge success probability, and local success probability are determined based on the edge server task information, the bit length information, the delay constraint information, the energy constraint information, the transmission characteristic information, the edge characteristic information, and the local characteristic information. Based on the transmission success probability, the edge success probability, and the local success probability, a transmit power optimization subtask, a transmission duration optimization subtask, and a task segmentation ratio optimization subtask are determined. The transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask are iteratively solved to determine the resource allocation probability of the target system.

[0008] In one embodiment, the step of determining the transmission success probability, edge success probability, and local success probability based on the edge server task information, the bit length information, the delay constraint information, the energy constraint information, the transmission characteristic information, the edge characteristic information, and the local characteristic information includes: The edge server share and local server share are determined based on the edge server task information. The probability of successful transmission is calculated based on the edge server share, the bit length information, the bandwidth, server-allocated transmission duration, transmission power, channel range, and noise power in the transmission characteristic information. The edge success probability is calculated based on the edge server share, the bit length information, the feasibility constraints in the edge feature information, the bit required period, and the cumulative transmission delay. The local success probability is calculated based on the latency constraint information, the energy constraint information, the transmission power in the local feature information, and the transmission duration allocated by the server.

[0009] In one embodiment, the step of determining the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask based on the transmission success probability, the edge success probability, and the local success probability includes: The overall success probability and the interruption probability are calculated based on the transmission success probability, the edge success probability, and the local success probability. The joint optimization solution task is determined based on the overall success probability and the interruption probability; The joint optimization solution task is decomposed into a transmit power optimization subtask, a transmission duration optimization subtask, and a task partitioning ratio optimization subtask.

[0010] In one embodiment, the step of iteratively solving the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask to determine the target system resource allocation probability includes: The transmit power optimization subtask is solved iteratively using Newton's method to determine the first solution result; The transmission duration optimization subtask is iteratively solved using the convex optimization method to determine the second solution result. The task segmentation ratio optimization subtask is iteratively solved using the successive upper bound minimization method to determine the third solution result. The resource allocation probability of the target system is determined based on the first solution result, the second solution result, and the third solution result.

[0011] In one embodiment, the user quantity information is multiple users; The step of determining the resource allocation probability of the target system based on the edge server task information, the bit length information, the latency constraint information, and the energy constraint information includes: Acquire information on interference between users, transmission time allocation between users, and user characteristics; The task segmentation ratio is determined based on the edge server task information; The probability of successful transmission between users and the probability of successful edge transmission between users are calculated based on the task segmentation ratio, the interference between users, the transmission time allocated between users, the bit length information, the delay constraint information, the energy constraint information, the transmission duration between users and the transmission power between users in the user characteristic information. The joint optimization solution task between users is determined based on the success probability of transmission between users and the success probability of edge connection between users. The joint optimization task among users is solved iteratively using deep reinforcement learning methods to obtain the resource allocation probability of the target system.

[0012] In one embodiment, the step of iteratively solving the joint optimization task among users using a deep reinforcement learning method to obtain the resource allocation probability of the target system includes: The joint optimization task among users is decomposed into multiple Markov decision subtasks using deep reinforcement learning methods, which determine the state space task, action space task, reward function task, and state transition task. The resource allocation probability of the target system is obtained by iteratively solving the following components: task size vector, channel state information, edge server queue state, user energy level, task segmentation ratio matrix, transmission duration matrix, transmit power vector, energy capacity and energy efficiency weight in the reward function task, and channel change component and task arrival component in the state transition task.

[0013] Furthermore, to achieve the above objectives, this application also proposes an over-the-air edge computing resource allocation device oriented towards execution uncertainty, the over-the-air edge computing resource allocation device oriented towards execution uncertainty includes: The acquisition module is used to acquire user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; The processing module is used to determine the resource allocation probability of the target system based on the user number information, the edge server task information, the bit length information, the delay constraint information, and the energy constraint information. The execution module is used to allocate resources based on the target system resource allocation probability control system, and to complete the allocation of airborne edge computing resources in the face of execution uncertainty.

[0014] Furthermore, to achieve the above objectives, this application also proposes an over-the-air edge computing resource allocation device for execution uncertainty, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the over-the-air edge computing resource allocation method for execution uncertainty as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the above-described method for allocating airborne edge computing resources oriented to execution uncertainty.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This embodiment proposes a method for allocating resources in over-the-air edge computing oriented towards execution uncertainty. It acquires user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information. Based on these information, it determines the resource allocation probability of the target system. The system then uses this probability to control resource allocation, thus completing the allocation of over-the-air edge computing resources oriented towards execution uncertainty. By acquiring user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information, this application quantifies uncertainties in a dynamic environment into probabilistic constraints. It generates optimal resource allocation strategies using algorithms corresponding to different scenarios, calculates the resource allocation probability of the target system, and enables the system to adaptively allocate resources. This significantly improves the reliability of task completion under strict latency and energy constraints, maximizing system resource utilization efficiency. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an embodiment of the airborne edge computing resource allocation method for execution uncertainty in this application. Figure 2 This application presents a technical roadmap for a single-user AirComp system scenario regarding the airborne edge computing resource allocation method for performance uncertainty. Figure 3 This is a schematic diagram illustrating the application curves of the interruption probability and iteration count of the airborne edge computing resource allocation method for execution uncertainty in this application. Figure 4This is a schematic diagram of the convergence curves of DQN, the airborne edge computing resource allocation method for execution uncertainty in this application, at different learning rates. Figure 5 This is a schematic diagram comparing the DQN and BCD methods for allocating airborne edge computing resources in response to execution uncertainty, as presented in this application. Figure 6 Jane's Fairness Index for different user priority weighting ratios in the airborne edge computing resource allocation method for execution uncertainty in this application; Figure 7 This is a schematic diagram illustrating the decision-making time of different system scales for the airborne edge computing resource allocation method for execution uncertainty in this application. Figure 8 This is a flowchart illustrating Embodiment 2 of the method for allocating airborne edge computing resources to address execution uncertainty in this application. Figure 9 This application presents a technology roadmap for a multi-user AirComp system with an airborne edge computing resource allocation method for performance uncertainty. Figure 10 This is a schematic diagram of the module structure of an over-the-air edge computing resource allocation device for execution uncertainty, as described in an embodiment of this application. Figure 11 This is a schematic diagram of the device structure of the hardware operating environment involved in the airborne edge computing resource allocation method for execution uncertainty in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application embodiment is: to obtain user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; to determine the target system resource allocation probability based on the user quantity information, the edge server task information, the bit length information, the latency constraint information, and the energy constraint information; and to control the resource allocation based on the target system resource allocation probability to complete the allocation of airborne edge computing resources oriented towards execution uncertainty.

[0024] In this embodiment, for ease of description, the following description will focus on identifying an airborne edge computing resource allocation device that is subject to execution uncertainty.

[0025] While existing technologies in single-user scenarios model the number of CPU cycles required per bit as a random variable and attempt to jointly optimize related parameters, they fail to effectively integrate with multi-user interference and system dynamics, optimize communication and computation metrics separately, or consider randomness only at the single-user level. They fail to perform joint modeling under a unified probabilistic objective of timely completion, making it difficult to achieve global optimality under the dual constraints of latency and energy consumption. In multi-user scenarios, although methods such as deep reinforcement learning have begun to be used for online scheduling, they lack accurate probabilistic characterization and unified modeling of the inherent randomness of execution uncertainty, and the computational complexity is very high, making it impossible to achieve real-time adaptation.

[0026] This application provides a solution to obtain user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; determine the target system resource allocation probability based on the user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; and control the system based on the target system resource allocation probability to perform resource allocation, thereby completing the allocation of airborne edge computing resources oriented towards execution uncertainty.

[0027] As can be seen from the above embodiments, this application quantifies the uncertainties in the dynamic environment into probabilistic constraints by acquiring user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information. It then calculates the optimal resource allocation strategy generated by the corresponding algorithm under different scenarios to obtain the resource allocation probability of the target system. This enables the system to adaptively allocate resources, significantly improving the reliability of task completion under strict latency and energy constraints, and maximizing the system's resource utilization efficiency.

[0028] Based on this, embodiments of this application provide a method for allocating airborne edge computing resources to address execution uncertainties, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the airborne edge computing resource allocation method for performance uncertainty in this application.

[0029] In this embodiment, the method for allocating airborne edge computing resources to address execution uncertainty includes steps S10 to S30: Step S10: Obtain user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; It should be noted that the user quantity information refers to the number of mobile terminals participating in the computation in the system, which can be a single user or multiple users. The edge server task information refers to the allocation of user computation tasks on the edge server, including the task segmentation ratio, transmission duration, and server computing resources. The bit length information refers to the data size of the computation task, expressed in bits. The latency constraint information is the maximum allowable time limit for task completion, ensuring that the computation and transmission process is completed within a strict time limit. The energy constraint information is the energy budget limit of the user equipment, controlling the total energy consumption during task execution to not exceed the equipment capacity.

[0030] It is understandable that if the user count information indicates a single user, it represents a single-user scenario, where only one battery-limited mobile user interacts with multiple distributed edge servers. If the user count information indicates multiple users, it represents a multi-user scenario, where multiple users, such as N, need to simultaneously compete for shared wireless and computing resources. In a single-user scenario, the task allocation for the edge servers is determined by the user dividing the task into M+1 parts, where M is the number of edge servers, and the share allocated to edge server m is... The local share is And satisfy = 1, the share of the edge server task information allocated in a multi-user scenario is This represents the proportion of tasks allocated to edge server m, and satisfies... = 1, and the bit length information is denoted as L in a single-user scenario, representing the total number of bits in the task to be processed, and denoted as in a multi-user scenario. , representing the task size of the nth user, and the latency constraint information in a single-user scenario, denoted as . The unit is seconds. In multi-user scenarios, it is denoted as . , representing the latency constraint for the nth user, and the energy constraint information in a single-user scenario, denoted as . In multi-user scenarios, it is denoted as , where represents the energy limit of the nth user.

[0031] In a specific embodiment, upon system initialization or task arrival, key input information required for resource allocation can be dynamically obtained through a preset configuration interface, user device reporting, or environmental awareness module. User quantity information is confirmed by the access point through signaling interaction to determine the number of currently active terminals, distinguishing between single-user and multi-user working modes. Edge server task information originates from the task scheduler, specifying the pre-allocation strategy for tasks to be processed within the available edge server cluster, including a task splitting ratio vector. = [ 0, 1, ..., m ](in 0 represents the locally calculated share. m (The share allocated to the m-th edge server) and the TDMA transmission duration allocated to each server. Bit length information is provided directly by the task source, such as an application or sensor, and represents the total data size of the computing task to be processed. Latency constraint information can be set by real-time business requirements, specifying the maximum time threshold from task initiation to completion. Energy constraint information can be read from the battery management unit of the terminal device or the preset energy consumption budget, limiting the maximum energy value allowed to be consumed throughout the entire task execution process.

[0032] Step S20: Determine the target system resource allocation probability based on the user quantity information, the edge server task information, the bit length information, the latency constraint information, and the energy constraint information; It should be noted that the target system resource allocation probability is the overall probability of the system successfully completing the entire computation task under given time and energy constraints.

[0033] In a specific embodiment, taking a single application scenario as an example, at this time, such as Figure 2 As shown, Figure 2 This application presents a technical roadmap for a single-user AirComp system scenario, focusing on the airborne edge computing resource allocation method for performance uncertainty. A single mobile user, limited by battery capacity, wishes to complete a computationally intensive application using m distributed edge servers. This application requires strict latency constraints. Processing a data computation task containing L bits within seconds requires intelligent resource allocation across multiple computing nodes. In this case, transmission characteristic information, edge characteristic information, and local characteristic information are acquired. Based on the edge server task information, the share allocated to edge servers and the share allocated to local servers are determined. That is, the user divides the task into M+1 parts, where M is the number of edge servers, and the share allocated to edge server m is... The local share is And satisfy = 1.

[0034] Based on the edge server share, the bit length information, the bandwidth in the transmission characteristic information, the server-allocated transmission duration, the transmission power, the channel range, and the noise power, the transmission success probability is calculated. That is, TDMA sequential offloading is used, and the transmission duration allocated to server m is... The channel exhibits flat Rayleigh fading. , The boundary value is the bandwidth. The transmission power is The noise power is Under TDMA, the cumulative transmission delay of the m-th server is = .

[0035] Success rate of transmission It can be represented as:

[0036] at this time, = / χ(x,y) is a predefined constraint function.

[0037] CPU cycles required per bit Gamma(α, β) (independent and identically distributed).

[0038] The edge success probability is calculated based on the edge server share, the bit length information, the feasibility constraints in the edge feature information, the required bit period, and the cumulative transmission delay. The local success probability is calculated based on the delay constraint information, the energy constraint information, the transmission power in the local feature information, and the server-allocated transmission duration. The overall success probability and the interruption probability are calculated based on the transmission success probability, the edge success probability, and the local success probability. The joint optimization solution task is determined based on the overall success probability and the interruption probability.

[0039] That is, calculating edge computing latency , is represented as:

[0040] in, Let m be the computing speed of the m-th edge server.

[0041] At this point, the total delay Represented as:

[0042] in, It is the total transmission time from the first server to the m-th server.

[0043] This allows us to calculate the marginal success probability. , is represented as:

[0044] The feasibility constraints are as follows:

[0045] Total latency It needs to be limited by time delay constraints Internally, and time delay constraints There are also separate feasibility constraints. Additionally, energy constraints must be met. The total local computational energy consumption is calculated and expressed as follows:

[0046] Among them, transmission power It is shared, and the transmission duration allocated to server m is... The sum represents the transmission time of all edge servers. It refers to the local CPU speed. L is the number of random cycles processed locally. This indicates the local task volume.

[0047] At this point, the local success probability under time and energy constraints Represented as:

[0048] Where α and β are gamma distributions The shape and scale parameters of Gamma(α,β) are ρ, which represents the maximum effective computing resources that can be obtained by local computation under the dual constraints of latency and energy, in CPU cycles. ρ=min(A,B) indicates that successful local computation is limited by latency A and energy B, with the smaller constraint between A and B being used for the limitation.

[0049] At this point, the overall success probability With interruption probability They are represented as follows:

[0050] in, It is the probability of successful transmission. It is the probability of success at the margin. It represents the local success probability.

[0051] Therefore, the determined joint optimization task P1 can be expressed as:

[0052] Among them, the joint optimization solution task P1 includes multiple stochastic optimization problems. = [ 0, 1, ..., m And T = [T1, ..., T] m Directly solving problem P1 is challenging because the optimization variables in P1 are coupled with each other in the objective function and constraints. Therefore, the BCD method is needed to solve the following three subproblems P2-P4 alternately.

[0053] Decompose the joint optimization and solution task into a transmission power optimization subtask, a transmission duration optimization subtask, and a task segmentation ratio optimization subtask. That is, decompose the joint optimization and solution task P1 into a transmission power optimization subtask P2, a transmission duration optimization subtask P3, and a task segmentation ratio optimization subtask P4.

[0054]

[0055] By iteratively solving these subproblems, we can obtain the local optimal solution of problem P1. The advantage of this method is that each subproblem has a simpler structure and is thus easier to handle.

[0056] Use Newton's method to iteratively solve the transmission power optimization subtask to determine the first solution result. That is, perform iterative solution for the transmission power optimization subtask P2. Affected by the energy consumption constraint, ln Regarding is a piecewise function. Specifically, when < P, the energy consumption constraint is not activated, ρ = , and ln increases with , where:

[0057] is the solution of the following formula for :

[0058] When > P, the solution is:

[0059] Therefore, for the transmission power optimization subtask P2, the optimal * is:

[0060] Where:

[0061] According to the nature of the problem, F is a concave function with respect to , so Newton's method can be used to determine , thereby solving the transmission power optimization subtask P2.

[0062] Use convex optimization method to iteratively solve the transmission duration optimization subtask to determine the second solution result. That is, for the transmission duration optimization subtask P3, from the function nature, it can be known that ln is a concave function with respect to T, so P3 is directly a convex problem and can be directly solved by numerical software such as CVX.

[0063] The task partitioning ratio optimization subtask is iteratively solved using the successive upper bound minimization method to determine the third solution result. That is, for the task partitioning ratio optimization subtask P4, P4 optimizes the task partitioning between edge servers, but since the nonlinear function in the objective function is still nonconvex, a suboptimal solution based on quadratic approximation MM(MM2) is proposed.

[0064] A surrogate function for P4 is constructed using a quadratic approximation, making it more tractable while preserving convergence. The iterative process of the proposed MM2 is given below.

[0065] In the (L+1)th MM iteration, using:

[0066] Once the valid surrogate function for P4 is obtained, the problem of P4.1 can then be solved:

[0067] in This is the optimal solution to the problem in the Lth MM iteration (m∈{0,M}).

[0068] also:

[0069] and:

[0070] Therefore, problem P4.1 can be equivalent to problem P4.2, expressed as:

[0071] Where st represents the constraints, and μ is the dual variable. Given μ, the objective function is about... It can be separated and decomposed into problem P4.3:

[0072] And break it down into problem P4.4:

[0073]

[0074] For problem P4.3, a closed form can be obtained. Setting the derivative of the objective function to zero, we get:

[0075] Among them, g1 and g2 are both computational components.

[0076] Since the objective function is concave, there can be at most one root satisfying the constraints. If none of them are satisfied, then... and Substituting the objective function into the equation, we select the optimal solution. The closed-form solution to problem P4.4 can be obtained using the Ferrari method. Setting the first derivative of the objective function of problem P4.4 to zero yields a quartic equation, expressed as:

[0077] in:

[0078] Its four roots are:

[0079] in:

[0080]

[0081] Since the target is concave, there are at most one of four feasible roots; if none exists, then at the endpoints... , Substituting the values ​​into the objective function and selecting the optimal solution, the optimal solution to problem P4.4 can be obtained efficiently. , m∈M. Since problem P4.2 is convex, the optimal solution can be found using binary search. So that:

[0082] Then place = By solving problems P4.3 and P4.4 respectively, the suboptimal solution to P4 can be obtained through the MM2 algorithm.

[0083] Other MM schemes are also widely used in wireless communication. For example, the first-order approximation MM (MM1) is often used for DC problems. The P4 problem can be solved using MM1, based on a similar approach to MM2. However, because... Due to the presence of nonlinear terms, MM1 cannot provide a closed-form solution. Therefore, each iteration of BCD-MM requires a two-dimensional search using Newton's method and a binary search. In contrast, BCD-MM2 in this scheme provides a semi-closed-form result, requiring only one binary search per iteration.

[0084] The resource allocation probability of the target system is determined based on the first solution result, the second solution result, and the third solution result.

[0085] In one feasible implementation, step S20 may include steps A11 to A14: Step A11: Obtain transmission feature information, edge feature information, and local feature information; It should be noted that the transmission feature information is a set of parameters used to characterize the quality and capability of the wireless communication link from the user equipment to the edge server, the edge feature information is a set of parameters used to quantify the execution process of the task on the edge server and its uncertainties, and the local feature information is a set of parameters used to quantify the resource consumption and constraints involved when the task is executed locally on the user equipment.

[0086] It is understood that the transmission characteristic information can be physical layer parameters necessary for calculating the probability of successful transmission of a specific amount of data within a specific time period, such as system bandwidth, user transmit power, parameters of channel fading statistics, and receiver noise power. The edge characteristic information can be server capability and task attribute parameters necessary for calculating the probability that the task portion allocated to the edge server can be successfully completed within strict latency constraints, such as server computing speed, task ratio, parameters characterizing the randomness of computational complexity, and the cumulative transmission latency before the server starts computation due to serial offloading. The local characteristic information can be device capability and task attribute parameters necessary for calculating the probability that the local computation portion can be successfully completed under the dual constraints of latency and energy, such as local computation speed, energy efficiency coefficient of the local chip, local computational complexity parameters, local task ratio, and the energy remaining for local computation after deducting the total transmission energy consumption from the total energy.

[0087] Step A12: Determine the transmission success probability, edge success probability, and local success probability based on the edge server task information, the bit length information, the delay constraint information, the energy constraint information, the transmission characteristic information, the edge characteristic information, and the local characteristic information. It should be noted that the transmission success probability is the probability that the user equipment can successfully deliver its assigned task data without errors within the transmission time allocated to the edge server. The edge success probability is the probability that, after the user's task data is successfully transmitted to the edge server, the computation process can be completed within the final time limit of the entire task, given the computing power of the server. The local success probability is the probability that the user equipment can complete its assigned computation task locally under the dual constraints of latency and energy.

[0088] In one feasible implementation, step A12 may include steps B11 to B14: Step B11: Determine the edge server share and local server share based on the edge server task information; It should be noted that the edge server share refers to the proportion of tasks allocated to edge servers after the complete computing task is divided, while the local server share refers to the proportion of tasks that are retained for computing on the user's mobile device under the same task division.

[0089] It is understood that the edge server share is a continuous variable between 0 and 1, which specifies what proportion of the total task data needs to be offloaded to a specific edge server for processing. The local server share is also a variable between 0 and 1, representing the amount of task that the user chooses to process themselves. The allocation of this share is directly related to the latency and energy consumption generated by local computing, and is used to balance the offloading overhead and local resource consumption in resource allocation.

[0090] Step B12: Calculate the probability of successful transmission based on the edge server share, the bit length information, the bandwidth in the transmission feature information, the server-allocated transmission duration, the transmission power, the channel range, and the noise power. It is understood that the transmission success probability is a communication reliability indicator used to quantify the likelihood of a single uplink transmission success under random channel fading and noise interference.

[0091] Step B13: Calculate the edge success probability based on the edge server share, the bit length information, the feasibility constraints in the edge feature information, the bit required period, and the cumulative transmission delay; It is understood that the edge success probability is a computational reliability metric used to characterize the likelihood of success under execution uncertainty. For example, since the number of CPU cycles required per bit is a random variable, the computation completion time also varies randomly.

[0092] Step B14: Calculate the local success probability based on the delay constraint information, the energy constraint information, the transmission power in the local feature information, and the server-allocated transmission duration.

[0093] It is understood that the local success probability is a resource constraint satisfaction index, used to characterize the final success probability of a resource-constrained device executing a task locally, that is, whether the comprehensive local computing speed is sufficient to complete the calculation within the time limit, and whether the remaining energy after deducting all transmission energy consumption is sufficient to support high-power local computing.

[0094] Step A13: Determine the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask based on the transmission success probability, the edge success probability, and the local success probability. It should be noted that the transmit power optimization subtask is an independent task that specifically optimizes the transmit power of user equipment under the premise of fixed task segmentation ratio and transmission duration. The transmission duration optimization subtask is an independent task that specifically optimizes the transmission time vector allocated to each edge server under the premise of fixed transmit power and task segmentation ratio. The task segmentation ratio optimization subtask is an independent task that specifically optimizes the task segmentation ratio vector under the premise of fixed transmit power and transmission duration.

[0095] In one feasible implementation, step A13 may include steps C11 to C13: Step C11: Calculate the overall success probability and the interruption probability based on the transmission success probability, the edge success probability, and the local success probability; It should be noted that the overall success probability is the overall probability that the entire computing task is successfully executed under the specified time and energy constraints. The interruption probability is a complementary concept to the overall success probability, representing the probability that the entire system cannot successfully complete the task within the time and energy budget, that is, the probability that the system will experience a service interruption.

[0096] It is understandable that local execution and execution on each edge server are independent events. Therefore, the overall success probability is an end-to-end performance metric that takes a global perspective and integrates communication reliability and computational reliability. It quantifies the ultimate probability that the system can meet service requirements under the actual conditions of simultaneously considering the randomness of wireless channels, the randomness of task computational complexity, and tight resource constraints. The causes of interruption can be multiple, such as transmission failure in any link, computation timeout of any edge server or local device, or total energy consumption exceeding the budget. The interruption probability can directly characterize the risk of service failure. In the optimization process, maximizing the overall success probability is equivalent to minimizing the interruption probability. The interruption probability is reduced by optimizing resource allocation.

[0097] Step C12: Determine the joint optimization solution task based on the overall success probability and the interruption probability; It should be noted that the joint optimization task is a core mathematical optimization problem established to solve the resource allocation problem under execution uncertainty. That is, under the constraints of time delay and energy, it maximizes the overall success probability of the system by simultaneously optimizing three variables: the task segmentation ratio vector, the transmission duration vector, and the transmission power. These variables are highly coupled in the objective function and constraints. For example, the transmission power affects both the transmission success probability and the remaining computing energy, the transmission duration affects the cumulative delay and thus compresses the computing time, and the task segmentation ratio directly determines the distribution of communication and computing load. Directly solving this complex nonlinear and nonconvex problem is extremely difficult. Therefore, this application adopts the overall concept of block coordinate descent (BCD) to decompose this huge joint optimization task into three simpler and easier-to-handle subtasks for iterative solving, thereby effectively reducing the computational complexity while ensuring performance.

[0098] Step C13: Decompose the joint optimization solution task into a transmit power optimization subtask, a transmission duration optimization subtask, and a task segmentation ratio optimization subtask.

[0099] Understandably, the core objective of the transmit power optimization subtask is to balance transmission reliability and energy consumption, thereby significantly improving the probability of successful transmission. However, this will drastically consume the system's total energy budget, compress the energy available for local computation, and reduce the probability of local success. The core objective of the transmission duration optimization subtask is to rationally allocate limited time resources within strict total latency constraints. Increasing the computation time of any server can certainly improve its corresponding transmission success probability, but it will cumulatively increase the waiting time of subsequent servers, thereby compressing their computation time and reducing their computation success probability. The task partitioning ratio optimization subtask is necessary because variables appear in the success probability formulas of transmission, edge computing, and local computing in a complex nonlinear form, causing the computation function to be nonconvex. Therefore, it is necessary to construct an easily tractable surrogate function to approximate the original complex objective function, transforming the nonconvex problem into a series of decomposable convex subproblems, thereby efficiently obtaining high-quality suboptimal solutions and determining the optimal task offloading and local computing allocation strategy.

[0100] Step A14: Iteratively solve the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask to determine the target system resource allocation probability.

[0101] It is understandable that the target system resource allocation probability is a comprehensive core indicator that integrates the uncertainties in the communication and computing processes. It can be calculated based on the calculated local computing success probability, transmission success probability, and edge success probability.

[0102] In one feasible implementation, step A14 may include steps D11 to D14: Step D11: Iteratively solve the transmit power optimization subtask using Newton's method to determine the first solution result; It should be noted that the first solution is the optimal solution obtained by iteratively executing the transmit power optimization subtask.

[0103] Understandably, the first solution is the optimal power value that maximizes the overall success probability, obtained by analyzing the piecewise impact of energy consumption constraints and using Newton's method to numerically solve the problem under the condition of fixed task segmentation ratio and transmission duration in the current iteration. This represents the optimal communication energy control strategy obtained by independently optimizing the power variable in the current iteration round, and is fixed as a known condition for subsequent solutions to the transmission duration subtask and task segmentation ratio subtask, thereby gradually advancing the entire joint optimization problem toward the optimal solution.

[0104] Step D12: Iteratively solve the transmission duration optimization subtask using the convex optimization method to determine the second solution result; It should be noted that the second solution is the optimal solution obtained by iteratively executing the transmission duration optimization subtask, assuming that the first solution has been determined and fixed.

[0105] Understandably, the second solution result is the optimal transmission duration vector. Under the given power and task allocation scheme, by solving the convex optimization problem, an optimal transmission time is allocated to each edge server, thereby maximizing the overall success probability of the system.

[0106] Step D13: Iteratively solve the task segmentation ratio optimization subtask using the successive upper bound minimization method to determine the third solution result; It should be noted that the third solution is the optimal solution obtained by optimizing the subtasks by performing task partitioning ratio, provided that the first and second solutions have been determined and fixed.

[0107] It is understandable that the third solution result is the optimal task segmentation ratio vector. It is a high-quality suboptimal solution obtained by using the successive upper bound minimization algorithm to solve the non-convex problem. That is, a surrogate function is constructed to approximate the original complex goal, and the quartic equation and binary search are used to update the dual variables, thereby efficiently determining how to optimally allocate tasks between the local server and various edge servers to maximize the overall success probability.

[0108] Step D14: Determine the resource allocation probability of the target system based on the first solution result, the second solution result, and the third solution result.

[0109] It is understandable that by utilizing the first, second, and third solution results, the resource allocation probability of the target system can be maximized, thereby systematically ensuring the reliability and success rate of the service under the randomness of task execution and the tight constraints of resources.

[0110] Step S30: Based on the target system resource allocation probability control system, resource allocation is performed to complete the allocation of airborne edge computing resources for execution uncertainty.

[0111] It is understood that the resource allocation is the process by which the system transforms the target system resource allocation probability, which is the corresponding optimal task segmentation ratio, optimal transmission duration and optimal transmission power, obtained by the optimization algorithm into executable physical layer and control layer instructions, thereby coordinating and precisely configuring the three core resources between user equipment and edge server: computing tasks, wireless channel time slots and terminal energy.

[0112] In a specific embodiment, the computing load is split and distributed between the local server and multiple edge servers according to the optimal task splitting ratio vector; strict time-division multiple access timing scheduling is performed according to the optimal transmission duration vector, a dedicated transmission window is allocated to each server, and the power amplifier of the user equipment radio frequency unit is dynamically adjusted according to the optimal transmit power to keep it stable at the power level in all transmission periods. Under the dual constraints of latency and energy, the entire system is driven to maximize the probability of successful task execution, and to achieve adaptive and collaborative allocation of communication, computing and energy resources in an uncertain environment.

[0113] In single-user static parameter scenarios, it provides provably convergent, low-complexity optimal / suboptimal solutions; in multi-user dynamic and uncertain scenarios, it offers highly successful, low-latency, strongly fair, and energy-efficient adaptive strategies, reducing deployment costs and improving overall service quality and resource utilization. It features rapid convergence and low complexity, facilitating engineering deployment. Figure 3 As shown, Figure 3 This diagram illustrates the application curves of the outage probability versus iteration count for the airborne edge computing resource allocation method addressing execution uncertainty, as presented in this application. The diagram shows that BCD-MM2, based on a second-order approximation, converges within no more than 7 iterations in all test scenarios, achieving the same optimal outage probability as the first-order BCD-MM1, while significantly reducing the solution complexity per iteration. It is suitable for online / quasi-online edge computing, possesses scalability and a high success rate in its learning paradigm, and is suitable for edge-side online / quasi-online computing. Figure 4 As shown, Figure 4 This diagram illustrates the convergence curves of the DQN (Discretionary Qualification) method for allocating resources in airborne edge computing with varying learning rates, as described in this application. The DQN framework maintains an average success probability of ≥ 80% as the user base scales. Figure 5 As shown, Figure 5 This diagram illustrates the comparison between DQN and BCD, the over-the-air edge computing resource allocation methods proposed in this application for handling execution uncertainties. The traditional optimization baseline rapidly degrades to <40% with increasing user numbers; meanwhile, DQN training converges stably after approximately 800 iterations, balancing convergence speed and stability in its learning rate, demonstrating scalable and deployable learning capabilities, while also considering fairness and differentiated services. Figure 6 As shown, Figure 6 The figure shows the J.A. Fairness Index (FQI) for different user priority weight ratios in the proposed over-the-air edge computing resource allocation method for execution uncertainty. Under different priority weight ratios, DQN consistently maintains an FQI > 0.8. Compared to the significant imbalance in traditional scheduling methods such as weighted / proportional scheduling at high weight ratios, the proposed method can maintain high fairness while ensuring differentiated services. Furthermore, it features millisecond-level decision latency, meeting the needs of real-time services, such as… Figure 7 As shown, Figure 7 This diagram illustrates the decision-making time of the airborne edge computing resource allocation method for execution uncertainty in this application at different system scales. In large-scale systems (such as 12 users × 10 servers), DQN inference decision-making takes only 5–14 ms, which is significantly lower than BCD-MM (>200 ms) and gradient descent (≈130 ms), thus meeting the real-time requirements of edge intelligence.

[0114] This embodiment proposes an over-the-air edge computing resource allocation method oriented towards execution uncertainty. It acquires user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information. Based on these information, it determines the resource allocation probability of the target system. The system then uses this probability to control resource allocation, thus completing the over-the-air edge computing resource allocation oriented towards execution uncertainty. This solves the technical problem of overcoming task execution uncertainty in over-the-air edge computing to achieve efficient resource allocation in single-user and multi-user scenarios under latency and energy constraints. Compared to existing technologies, this application quantifies uncertainties in the dynamic environment into probabilistic constraints by acquiring user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information. It then uses algorithms corresponding to different scenarios to generate optimal resource allocation strategies for calculation, obtaining the resource allocation probability of the target system. This allows the system to adaptively allocate resources, significantly improving the reliability of task completion under strict latency and energy constraints and maximizing system resource utilization efficiency.

[0115] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the first embodiment can be referred to the above description, and will not be repeated hereafter.

[0116] In this embodiment, refer to Figure 8 , Figure 8 This is a flowchart illustrating Embodiment 2 of the method for allocating airborne edge computing resources to address execution uncertainty, and step S20 further includes steps S21 to S25: Step S21: Obtain information on interference between users, the allocation of transmission time between users, and user characteristics; It is understood that the inter-user interference refers to the sum of co-channel interference caused to the user's signal at the receiving end of the edge server by all other users transmitting simultaneously, except for the target user, when multiple users share the same wireless channel resources. The inter-user allocated transmission time is a dedicated time slot length allocated to each user in a shared, limited wireless resource pool for transmitting data to the edge server. The inter-user characteristic information is a set of unique and dynamically changing state parameters for each user, used to accurately describe the real-time status of each individual in the multi-user system.

[0117] In a specific embodiment, taking a multi-user scenario as an example, the interference between users is estimated in real time by the physical layer of each edge server through continuous measurement of the total received power of non-target user signals in the current time slot. The transmission time allocated between users is directly generated based on the resource allocation strategy, and the characteristic information between users may include the transmission duration between users and the transmission power between users.

[0118] Step S22: Determine the task segmentation ratio based on the edge server task information; It should be noted that the task splitting ratio refers to the ratio of local tasks to edge server tasks for each user in a multi-user scenario. The local task ratio indicates what proportion of tasks a user keeps for local computation, while the edge server task ratio indicates what proportion of their tasks are offloaded to edge servers for computation, thereby achieving computational load balancing in a multi-user competitive environment.

[0119] In a specific embodiment, taking a multi-user scenario as an example, the single-user AirComp system model is extended to N mobile users running simultaneously. Each user n∈N = 1, 2, ..., N needs to meet latency constraints. and energy budget Internal completion size is The computation task of bits is performed using the same M distributed edge servers. Following the previously introduced task allocation strategy, user n divides their computation task into M + 1 subtasks, where... This represents the proportion of tasks allocated to edge server m, i.e., the task partitioning ratio, and satisfies... = 1.

[0120] Step S23: Calculate the inter-user transmission success probability and inter-user edge success probability based on the task segmentation ratio, inter-user interference, inter-user allocated transmission time, bit length information, delay constraint information, energy constraint information, inter-user transmission duration and inter-user transmit power in the inter-user feature information. It is understood that the probability of successful transmission between users is the probability that a user can reliably deliver the task data assigned to it to the edge server within the transmission time allocated to it, under the real-world condition of co-frequency interference among multiple users. The probability of successful edge transmission between users is the probability that, under the premise that the computing resources of the edge server are shared by multiple user tasks, the task portion of the user offloaded to the server can still be completed within its own latency constraints after taking into account the computing load generated by other users.

[0121] In a specific embodiment, taking a multi-user scenario as an example, the probability of successful inter-user transmission when user n unloads to edge server m is... It can be represented as:

[0122] in, This represents the inter-user interference experienced by user n when transmitting data to edge server m. This is the transmission time allocated to user n.

[0123] At this point, user n calculates the success probability of edge connections between users on edge server m. It can be represented as:

[0124] The computing power of the edge server m needs to be shared among multiple users.

[0125] Step S24: Determine the joint optimization solution task between users based on the inter-user transmission success probability and the inter-user edge success probability; It should be noted that the joint optimization solution task among users is a core mathematical optimization problem that formally defines the resource allocation problem in a multi-user scenario. That is, under the premise of satisfying the task integrity constraints of each user, the total system transmission time constraints, and the computing power constraints of each edge server, the task segmentation ratio matrix, transmission power vector, and transmission time of all users are jointly and synchronously optimized to maximize the sum of the weighted success probabilities of all users.

[0126] It is understandable that the joint optimization task among users spans different users, and there is strong cross-user coupling in the objective function and constraints. For example, if one user increases its transmission power, it will exacerbate the interference to other users. If one user occupies too much transmission time or computing resources, it will directly compress the available resources of other users. Therefore, it is impossible to solve the problem using a single-user problem.

[0127] In a specific embodiment, such as Figure 9 As shown, Figure 9 This application presents a technical roadmap for a multi-user AirComp system with an airborne edge computing resource allocation method for execution uncertainty. The joint optimization solution task P5 among users, i.e., the multi-user success probability maximization problem, can be expressed as:

[0128]

[0129] in, This represents the priority weight of user n. Let n be the probability of success for user n. For the maximum time slot length, This represents the computing power constraint of the edge server m.

[0130] Step S25: Use deep reinforcement learning to iteratively solve the joint optimization task among users to obtain the resource allocation probability of the target system.

[0131] Understandably, for the aforementioned joint optimization task among users, the system can model the resource allocation problem as a Markov Decision Process (MDP). In this process, an agent observes the current system state in each time slot. The agent can be a resource scheduler, and the current system state can include channel conditions, task queues, user energy, etc. Then, based on its policy network, the agent selects an action—a specific resource allocation scheme—including task splitting, transmission time, and power. The environment then transitions to the next state and provides a reward signal. This reward comprehensively reflects the weighted success probability and energy efficiency. Through algorithms such as Deep Q-Network (DQN) and with the aid of experience replay and target networks, the agent iteratively updates its Q-network parameters across numerous state-action-reward-new state trajectories. Ultimately, the policy converges to a state that adapts to dynamic environmental changes, thereby approximating the joint optimization task among users and generating the optimal or approximate target system resource allocation probability online.

[0132] In a specific embodiment, a deep reinforcement learning method is used to decompose the joint optimization task among users into multiple Markov decision subtasks, identifying the state space task, action space task, reward function task, and state transition task. Based on the task size vector, channel state information, edge server queue state, and user energy level in the state space task; the task segmentation ratio matrix, transmission duration matrix, and transmit power vector in the action space task; the energy capacity and energy efficiency weight in the reward function task; and the channel change component and task arrival component in the state transition task, an iterative solution is performed to obtain the resource allocation probability of the target system. That is, the multi-user resource allocation problem can be modeled as a Markov decision process, including four elements: state space, action space, reward function, and state transition.

[0133] Wherein, in the state space, the state of the system in time slot t is defined as:

[0134] in, =[ [Indicates task size] Includes channel state information. Indicates the queue status of the edge server. Indicates the user's energy level.

[0135] In the action space, if the action of the system in time slot t is defined as:

[0136] in, = , = , = .

[0137] In the reward function, the instant reward is designed to maximize the weighted sum of success probabilities while considering energy efficiency, and is expressed as:

[0138] in, As an energy efficiency weight, Let n be the maximum energy capacity for user n.

[0139] In the state transitions described above, the system state transitions follow the following dynamics:

[0140] in, This represents random factors such as channel changes and mission arrival.

[0141] A Deep Q-Network (DQN) approach is employed to solve the multi-user resource allocation problem. This algorithm maintains a deep neural network Q(s, a, θ) to approximate the optimal Q-function, where θ represents the network parameters. The Q-Network consists of multiple fully connected layers and uses the ReLU activation function. The input layer receives the system state vector, the hidden layers capture the complex relationships between system parameters, and the output layer generates the Q-values ​​for all possible actions. To handle the continuous action space, an action discretization method is used, quantizing continuous variables into a finite number of levels. For example, the task allocation ratio... Using step size Δ The uniform discretization is set to 0.1, and transmission time and transmit power are quantized with appropriate granularity. Furthermore, during training, experience replay, a target network, dual DQN, and prioritized experience replay are employed to improve convergence and stability. Experience replay uses a replay buffer to store historical experience, breaking temporal correlation and improving sample utilization. The target network provides a stable Q-value target during training using a separate target network. Dual DQN uses a main network for action selection and a target network for value evaluation to mitigate overestimation bias. Prioritized experience replay samples important transitions at a higher frequency based on temporal difference errors.

[0142] In one feasible implementation, step S25 may include steps E11-E12: Step E11: The joint optimization solution task among users is decomposed into multiple Markov decision subtasks using deep reinforcement learning methods, and the state space task, action space task, reward function task, and state transition task are determined. Understandably, using deep reinforcement learning methods to decompose the joint optimization task among users into multiple Markov decision subtasks, namely, the state space task, which requires designing a feature vector that can comprehensively capture the real-time dynamics of the system, such as channel state, task queue, user energy, etc.; the action space task, which requires the reasonable discretization of continuous resource allocation decisions, such as task splitting ratio, transmission duration and transmission power, to form selectable actions; the reward function task, which requires designing an instantaneous reward signal that can simultaneously reflect the optimization objective and system constraints; and the state transition task, which requires modeling the state evolution law of the system after the execution of actions due to random changes in the channel and random arrival of tasks.

[0143] Step E12 involves iteratively solving for the target system resource allocation probability based on the task size vector, channel state information, edge server queue state, user energy level, task segmentation ratio matrix, transmission duration matrix, transmit power vector, energy capacity, energy efficiency weight, channel change component, and task arrival component in the state transition task within the state space task.

[0144] It is understandable that the network parameters can be converged to a stable state through iterative solutions, thereby outputting a resource allocation action that can approximately maximize the long-term cumulative reward for any given system state in real time, and obtaining the final target system resource allocation probability.

[0145] This embodiment proposes an over-the-air edge computing resource allocation method for execution uncertainty. The method acquires inter-user interference, inter-user allocated transmission time, and inter-user characteristic information. It determines a task segmentation ratio based on the edge server task information. Based on the task segmentation ratio, inter-user interference, inter-user allocated transmission time, bit length information, delay constraint information, energy constraint information, and the inter-user transmission duration and inter-user transmit power from the inter-user characteristic information, it calculates the inter-user transmission success probability and the inter-user edge success probability. Based on the inter-user transmission success probability and the inter-user edge success probability, it determines a joint optimization solution task. Finally, it uses a deep reinforcement learning method to iteratively solve the joint optimization solution task to obtain the target system resource allocation probability. This paper addresses the technical challenge of overcoming task execution uncertainties in edge computing to achieve efficient resource allocation in single-user and multi-user scenarios under latency and energy constraints. Compared to existing technologies, this application precisely quantifies the probability of communication and computation uncertainties by calculating the success probability of transmission and edge success between users. This serves as the optimization objective, formalizing the complex multi-user resource allocation problem as a high-dimensional joint optimization task. A deep reinforcement learning framework is used to model this task as a Markov decision process, with continuous interaction and policy iteration. The system autonomously learns the optimal task segmentation, transmission scheduling, and power control strategies. Under the premise of strictly meeting the differentiated latency and energy consumption constraints of each user, this significantly improves the overall success probability of concurrent execution of multiple tasks, effectively reduces the probability of system interruption, and achieves efficient, fair, and adaptive allocation of computing and communication resources.

[0146] This application also provides an airborne edge computing resource allocation device for performance uncertainty; please refer to [reference needed]. Figure 10 The above-ground edge computing resource allocation device for execution uncertainty includes: The acquisition module 10 is used to acquire user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; Processing module 20 is used to determine the resource allocation probability of the target system based on the user number information, the edge server task information, the bit length information, the delay constraint information, and the energy constraint information; The execution module 30 is used to allocate resources based on the target system resource allocation probability control system to complete the allocation of airborne edge computing resources in the face of execution uncertainty.

[0147] The over-the-air edge computing resource allocation device for execution uncertainty provided in this application adopts the over-the-air edge computing resource allocation method for execution uncertainty in the above embodiments. It can solve the technical problem of how to overcome the uncertainty of task execution in over-the-air edge computing, so as to achieve efficient resource allocation in single-user and multi-user scenarios under latency and energy constraints. Compared with the prior art, the beneficial effects of the over-the-air edge computing resource allocation device for execution uncertainty provided in this application are the same as the beneficial effects of the over-the-air edge computing resource allocation method for execution uncertainty provided in the above embodiments, and other technical features in the over-the-air edge computing resource allocation device for execution uncertainty are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0148] This application provides an over-the-air edge computing resource allocation device for execution uncertainty. The over-the-air edge computing resource allocation device for execution uncertainty includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the over-the-air edge computing resource allocation method for execution uncertainty in the above embodiment 1.

[0149] The following is for reference. Figure 11 This document illustrates a structural schematic diagram of an over-the-air edge computing resource allocation device suitable for implementing embodiments of this application, oriented towards execution uncertainty. The over-the-air edge computing resource allocation device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 11 The illustrated over-the-air edge computing resource allocation device for execution uncertainty is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0150] like Figure 11As shown, an over-the-air edge computing resource allocation device for execution uncertainty may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes based on programs stored in ROM (Read Only Memory) 1002 or programs loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the over-the-air edge computing resource allocation device for execution uncertainty. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the over-the-air edge computing resource allocation device for performing uncertain tasks to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows an over-the-air edge computing resource allocation device for performing uncertain tasks with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0151] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0152] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0153] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for allocating airborne edge computing resources to address execution uncertainty, characterized in that, The method includes: Obtain user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; The target system resource allocation probability is determined based on the user quantity information, the edge server task information, the bit length information, the latency constraint information, and the energy constraint information. Resource allocation is performed based on the target system resource allocation probability control system to complete the allocation of airborne edge computing resources oriented towards execution uncertainty.

2. The method as described in claim 1, characterized in that, The user count information refers to a single user. The step of determining the resource allocation probability of the target system based on the edge server task information, the bit length information, the latency constraint information, and the energy constraint information includes: Acquire transmission feature information, edge feature information, and local feature information; The transmission success probability, edge success probability, and local success probability are determined based on the edge server task information, the bit length information, the delay constraint information, the energy constraint information, the transmission characteristic information, the edge characteristic information, and the local characteristic information. Based on the transmission success probability, the edge success probability, and the local success probability, a transmit power optimization subtask, a transmission duration optimization subtask, and a task segmentation ratio optimization subtask are determined. The transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask are iteratively solved to determine the resource allocation probability of the target system.

3. The method as described in claim 2, characterized in that, The steps of determining the transmission success probability, edge success probability, and local success probability based on the edge server task information, the bit length information, the delay constraint information, the energy constraint information, the transmission characteristic information, the edge characteristic information, and the local characteristic information include: The edge server share and local server share are determined based on the edge server task information. The probability of successful transmission is calculated based on the edge server share, the bit length information, the bandwidth, server-allocated transmission duration, transmission power, channel range, and noise power in the transmission characteristic information. The edge success probability is calculated based on the edge server share, the bit length information, the feasibility constraints in the edge feature information, the bit required period, and the cumulative transmission delay. The local success probability is calculated based on the latency constraint information, the energy constraint information, the transmission power in the local feature information, and the transmission duration allocated by the server.

4. The method as described in claim 2, characterized in that, The steps of determining the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask based on the transmission success probability, the edge success probability, and the local success probability include: The overall success probability and the interruption probability are calculated based on the transmission success probability, the edge success probability, and the local success probability. The joint optimization solution task is determined based on the overall success probability and the interruption probability; The joint optimization solution task is decomposed into a transmit power optimization subtask, a transmission duration optimization subtask, and a task partitioning ratio optimization subtask.

5. The method as described in claim 2, characterized in that, The step of iteratively solving the transmit power optimization subtask, the transmission duration optimization subtask, and the task segmentation ratio optimization subtask to determine the target system resource allocation probability includes: The transmit power optimization subtask is solved iteratively using Newton's method to determine the first solution result; The transmission duration optimization subtask is iteratively solved using the convex optimization method to determine the second solution result. The task segmentation ratio optimization subtask is iteratively solved using the successive upper bound minimization method to determine the third solution result. The resource allocation probability of the target system is determined based on the first solution result, the second solution result, and the third solution result.

6. The method as described in claim 1, characterized in that, The user count information indicates multiple users; The step of determining the resource allocation probability of the target system based on the edge server task information, the bit length information, the latency constraint information, and the energy constraint information includes: Acquire information on interference between users, transmission time allocation between users, and user characteristics; The task segmentation ratio is determined based on the edge server task information; The probability of successful transmission between users and the probability of successful edge transmission between users are calculated based on the task segmentation ratio, the interference between users, the transmission time allocated between users, the bit length information, the delay constraint information, the energy constraint information, the transmission duration between users and the transmission power between users in the user characteristic information. The joint optimization solution task between users is determined based on the success probability of transmission between users and the success probability of edge connection between users. The joint optimization task among users is solved iteratively using deep reinforcement learning methods to obtain the resource allocation probability of the target system.

7. The method as described in claim 6, characterized in that, The step of iteratively solving the joint optimization task among users using deep reinforcement learning methods to obtain the resource allocation probability of the target system includes: The joint optimization task among users is decomposed into multiple Markov decision subtasks using deep reinforcement learning methods, which determine the state space task, action space task, reward function task, and state transition task. The resource allocation probability of the target system is obtained by iteratively solving the following components: task size vector, channel state information, edge server queue state, user energy level, task segmentation ratio matrix, transmission duration matrix, transmit power vector, energy capacity and energy efficiency weight in the reward function task, and channel change component and task arrival component in the state transition task.

8. A device for allocating airborne edge computing resources to address execution uncertainty, characterized in that, The device includes: The acquisition module is used to acquire user quantity information, edge server task information, bit length information, latency constraint information, and energy constraint information; The processing module is used to determine the resource allocation probability of the target system based on the user number information, the edge server task information, the bit length information, the delay constraint information, and the energy constraint information. The execution module is used to allocate resources based on the target system resource allocation probability control system, and to complete the allocation of airborne edge computing resources in the face of execution uncertainty.

9. A device for allocating airborne edge computing resources to address execution uncertainty, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method for allocating airborne edge computing resources oriented to execution uncertainty as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps of the method for allocating airborne edge computing resources oriented to execution uncertainty as described in any one of claims 1 to 7.

Images