Edge network resource allocation method and apparatus, computer device, and medium

Abstract

The application provides an edge network resource allocation method and device, computer equipment and a medium, comprising: obtaining a user's computing request; based on the computing request, mapping the computing request to an identification space to obtain an identification space mapping model, the identification space mapping model being used to represent the mapping of the user and server relationship; based on the identification space mapping model, calculating the energy cost and time cost of a computing task corresponding to the computing request to construct a multi-objective optimization model; based on the identification space mapping model and the multi-objective optimization model, determining an optimal allocation strategy for the computing task. By constructing a system model of identification space mapping, the computing task of the edge network system is mapped to the identification space to obtain an optimization model of the energy cost and time cost of the edge network system, the optimization model is used to adjust and optimize the resource arrangement strategy, and the resource allocation strategy with the minimum energy cost and time cost is obtained, thereby ensuring the reliability and network performance of the edge network system.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) technology, specifically to edge network resource allocation methods, devices, computer equipment, and media. Background Technology

[0002] Currently, the surge in data-intensive IoT services, such as industrial automation, smart healthcare, autonomous driving, and remote monitoring, is driving the rapid growth of data traffic in the Industry 4.0 era. Therefore, the IoT needs powerful analytics and computing capabilities to handle these computationally intensive tasks. To adapt to the growing demands of intelligence and digitalization, and by incorporating advanced artificial intelligence mechanisms, the IoT needs to analyze massive amounts of data to detect, identify, and discover environmental conditions. However, the AI-cloud IoT model still faces some challenges in terms of time sensitivity, resource utilization, and reliability.

[0003] In related technologies, the multi-access edge computing (MEC) paradigm has been envisioned as a potential candidate for providing highly available network resources at the IoT receiver layer. However, current IoT resource allocation methods cannot adapt to the explosive growth of data and suffer from significant edge resource consumption, resulting in IoT MEC networks still experiencing immense resource consumption pressure. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the shortcomings of existing Internet of Things resource allocation methods, which cannot adapt to explosive data growth and have large edge resource consumption, thereby providing edge network resource allocation methods, devices, computer equipment and media.

[0005] In conjunction with the first aspect, the present invention provides an edge network resource allocation method, the method comprising:

[0006] Obtain the user's computation request;

[0007] Based on the computation request, the computation request is mapped to the identifier space to obtain the identifier space mapping model, which is used to represent the mapping of the relationship between the user and the server;

[0008] Based on the identifier space mapping model, the energy cost and time cost of the computing task corresponding to the computing request are calculated, and a multi-objective optimization model is constructed.

[0009] Based on the identifier space mapping model and the multi-objective optimization model, the optimal allocation strategy for the computing task is determined.

[0010] In this approach, a system model mapping an identifier space is constructed to map the computational tasks of the edge network system to the identifier space, resulting in an optimization model for the energy and time costs of the edge network system. This optimization model is then used to adjust and optimize resource orchestration strategies, yielding a resource allocation strategy that minimizes both energy and time costs, thus supporting efficient and reliable resource allocation. By considering the multi-dimensional attribute values ​​of terminals and services, the reliability and network performance of the edge network system are simultaneously guaranteed. Furthermore, by matching the relationship between supply and demand, the reliability, energy consumption, and power consumption of the edge network system are jointly optimized.

[0011] In conjunction with the first aspect, in the first embodiment of the first aspect, the identifier space includes an access space and a matching space;

[0012] The step of mapping the computation request to the identifier space mapping model based on the computation request includes:

[0013] Based on the computing request, a computing task is determined, which includes a computing unloading task and a resource allocation task.

[0014] The computational offloading task is mapped to the access space to obtain the access mapping matrix;

[0015] The resource allocation task is mapped to the matching space to obtain the matching mapping matrix;

[0016] The identifier space mapping model is obtained by matching the access mapping matrix with the mapping matrix.

[0017] In conjunction with the first embodiment of the first aspect, in the second embodiment of the first aspect, the step of calculating the energy cost and time cost of the computing task corresponding to the computing request based on the identifier space mapping model, and constructing a multi-objective optimization model, includes:

[0018] Based on the access mapping matrix, the energy cost and time cost of the user-end computing task in the computing task are calculated to obtain the user-end energy cost and time cost optimization problem.

[0019] Based on the matching mapping matrix, the energy cost and time cost of the server-side computing task in the computing task are calculated to obtain the server-side energy cost and time cost optimization problem.

[0020] Based on the optimization problems of energy cost and time cost on the user side and the optimization problems of energy cost and time cost on the server side, a multi-objective optimization model is constructed.

[0021] In conjunction with the second embodiment of the first aspect, in the third embodiment of the first aspect, determining the optimal allocation strategy for the computational task based on the identifier space mapping model and the multi-objective optimization model includes:

[0022] Based on the access mapping matrix, the resource supply and demand relationship between the server and the user is determined;

[0023] Based on the resource supply and demand relationship between the server and the user, the optimal matching position is determined;

[0024] Based on the multi-objective optimization model and the optimal matching position, a matching strategy optimization model is constructed;

[0025] Based on the matching strategy optimization model, the optimal allocation strategy for the computing task is determined.

[0026] In conjunction with the third embodiment of the first aspect, in the fourth embodiment of the first aspect, determining the resource supply and demand relationship between the server and the user based on the access mapping matrix includes:

[0027] Obtain a supply and demand relationship model;

[0028] By inputting the access mapping matrix into the supply and demand relationship model, the resource supply and demand relationship between the server and the user is obtained.

[0029] In conjunction with the fourth embodiment of the first aspect, in the fifth embodiment of the first aspect, the supply and demand relationship model is a pre-trained deep neural network matching model.

[0030] In conjunction with the third embodiment of the first aspect, in the sixth embodiment of the first aspect, determining the optimal allocation strategy for the computational task based on the matching strategy optimization model includes:

[0031] Based on the multi-objective optimization model, determine the initial matching strategy;

[0032] Based on the matching strategy optimization model, the initial matching strategy is adjusted to minimize the energy and time costs of the system, thereby obtaining the optimal allocation strategy for the computing task.

[0033] In a second aspect, the present invention also provides an edge network resource allocation apparatus, the apparatus comprising:

[0034] The acquisition unit is used to acquire the user's computing request;

[0035] A mapping unit is used to map the computation request to an identifier space based on the computation request, thereby obtaining an identifier space mapping model, which is used to represent the mapping of the relationship between the user and the server.

[0036] The computing unit is used to calculate the energy cost and time cost of the computing task corresponding to the computing request based on the identifier space mapping model, and to construct a multi-objective optimization model.

[0037] The determining unit is used to determine the optimal allocation strategy for the computing task based on the identifier space mapping model and the multi-objective optimization model.

[0038] In conjunction with the second aspect, in the first embodiment of the second aspect, the identifier space includes an access space and a matching space;

[0039] The mapping unit includes:

[0040] The first determining unit is configured to determine a computing task based on the computing request, wherein the computing task includes a computing unloading task and a resource allocation task.

[0041] An access mapping unit is used to map the computational offloading task to the access space to obtain an access mapping matrix;

[0042] A matching mapping unit is used to map the resource allocation task to the matching space to obtain a matching mapping matrix;

[0043] The spatial mapping unit is used to obtain the identifier spatial mapping model by matching the mapping matrix based on the access mapping matrix.

[0044] In conjunction with the first embodiment of the second aspect, in the second embodiment of the second aspect, the computing unit includes:

[0045] The user-side computing unit is used to calculate the energy cost and time cost of the user-side computing task in the computing task based on the access mapping matrix, so as to obtain the user-side energy cost and time cost optimization problem.

[0046] The server-side computing unit is used to calculate the energy cost and time cost of the server-side computing task in the computing task based on the matching mapping matrix, so as to obtain the server-side energy cost and time cost optimization problem.

[0047] The first construction unit is used to construct a multi-objective optimization model based on the user-side energy cost and time cost optimization problem and the server-side energy cost and time cost optimization problem.

[0048] In conjunction with the second embodiment of the second aspect, in the third embodiment of the second aspect, the determining unit includes:

[0049] The supply and demand matching unit is used to determine the resource supply and demand relationship between the server and the user based on the access mapping matrix.

[0050] The location matching unit is used to determine the optimal matching location based on the resource supply and demand relationship between the server and the user.

[0051] The second construction unit is used to construct a matching strategy optimization model based on the multi-objective optimization model and the optimal matching position;

[0052] The strategy determination unit is used to determine the optimal allocation strategy for the computing task based on the matching strategy optimization model.

[0053] In conjunction with the third embodiment of the second aspect, in the fourth embodiment of the second aspect, the supply and demand matching unit includes:

[0054] The first acquisition unit is used to acquire the supply and demand relationship model;

[0055] The input unit is used to input the access mapping matrix into the supply and demand relationship model to obtain the resource supply and demand relationship between the server and the user.

[0056] In conjunction with the fourth embodiment of the second aspect, in the fifth embodiment of the second aspect, the supply and demand relationship model is a pre-trained deep neural network matching model.

[0057] In conjunction with the third embodiment of the second aspect, in the sixth embodiment of the second aspect, the strategy determination unit includes:

[0058] The first determining unit is used to determine the initial matching strategy based on the multi-objective optimization model;

[0059] The adjustment unit is used to adjust the initial matching strategy based on the matching strategy optimization model to minimize the energy cost and time cost of the system, thereby obtaining the optimal allocation strategy for the computing task.

[0060] According to a third aspect, embodiments of the present invention also provide a computer device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the edge network resource allocation method of any one of the first aspects and its optional embodiments.

[0061] According to a fourth aspect, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions for causing the computer to perform the edge network resource allocation method of the first aspect and any of its optional embodiments. Attached Figure Description

[0062] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0063] Figure 1 This is a flowchart of an edge network resource allocation method proposed according to an exemplary embodiment.

[0064] Figure 2 This is a schematic diagram of the structure of an edge network system according to an exemplary embodiment.

[0065] Figure 3 This is a structural block diagram of an edge network resource allocation device according to an exemplary embodiment.

[0066] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an exemplary embodiment. Detailed Implementation

[0067] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] In related technologies, the multi-access edge computing (MEC) paradigm has been envisioned as a potential candidate for providing highly available network resources at the IoT receiver layer. However, current IoT resource allocation methods cannot adapt to the explosive growth of data and suffer from significant edge resource consumption, resulting in IoT MEC networks still experiencing immense resource consumption pressure.

[0069] To address the aforementioned problems, this invention provides an edge network resource allocation method for use in a computer device. It should be noted that the executing entity can be an edge network resource allocation device, which can be implemented as part or all of the computer device through software, hardware, or a combination of both. The computer device can be a terminal, client, or server. The server can be a single server or a server cluster composed of multiple servers. In this embodiment, the terminal can be a smartphone, personal computer, tablet computer, wearable device, or other intelligent hardware device such as a smart robot. The following method embodiments all use a computer device as the executing entity for illustration.

[0070] The computer equipment in this embodiment is suitable for use scenarios in IoT edge networks. The edge network resource allocation method provided by this invention constructs a system model that maps the edge network system's computing tasks to the identifier space, obtaining an optimization model for the energy and time costs of the edge network system. This optimization model is then used to adjust and optimize resource orchestration strategies, resulting in a resource allocation strategy that minimizes both energy and time costs, thus supporting efficient and reliable resource allocation. By considering the multi-dimensional attribute values ​​of terminals and services, the reliability and network performance of the edge network system are simultaneously guaranteed. Furthermore, by matching the relationship between supply and demand, the reliability, energy consumption, and power consumption of the edge network system are jointly optimized.

[0071] Figure 1 This is a flowchart illustrating an edge network resource allocation method according to an exemplary embodiment. Figure 1 As shown, the edge network resource allocation method includes the following steps S101 to S104.

[0072] In step S101, the user's calculation request is obtained.

[0073] In this embodiment of the invention, the user's computation request is used to represent the user sending a request to the server to perform computation unloading.

[0074] Figure 2 This is a schematic diagram of the structure of an edge network system according to an exemplary embodiment. For example... Figure 2 As shown, in one example, the edge network system includes: S MEC servers MEC1Server to MEC... S Server, R gateways GW1 to GW r and K mobile devices MD1 to MD K The MEC servers are evenly distributed throughout the region. The server can establish wired connections with multiple gateways (GWs), and mobile devices... Access to a GW is achieved through a single antenna. To ensure reliable computational offloading, request data can reach the MEC server via multiple links. In this model, As a coordinator of resources and demands within the system, and a space converter, it determines the offloading location, the number of floating tasks per MD, and the allocation of computing resources to minimize the energy and time consumption of computing tasks. Once a computing task request arrives from a user, the selection of an offloading point is performed first, followed by resource allocation across the server cluster.

[0075] In step S102, based on the computation request, the computation request is mapped to the identifier space to obtain the identifier space mapping model.

[0076] In this embodiment of the invention, the identifier space mapping model is used to represent the mapping of the relationship between users and servers. The identifier space includes an access space and a matching space. Mapping a computation request to the identifier space mapping model based on the computation request includes: determining computation tasks based on the computation request, the computation tasks including computation unloading tasks and resource allocation tasks; mapping the computation unloading tasks to the access space to obtain an access mapping matrix; mapping the resource allocation tasks to the matching space to obtain a matching mapping matrix; and obtaining the identifier space mapping model based on the access mapping matrix and the matching mapping matrix.

[0077] In one example, the identifier space mapping model may include: access space (AS) k ) and matching space (MS k Among them, AS k ={AID k |ABD k}and AID K Representative of the requesting mobile device MD k The unique and legitimate access identifier, ABD k As a mobile device MD k The description of access behavior, i.e. CQ k For mobile device MD k Requested number of calculations; UL k For mobile device MD k User level, UL k ∈{UL1,...,UL N};SL k For mobile device MD k The service level (e.g., time-sensitive requirements) of the computational unloading task, SL k ∈{SL1,...,SL M};SR k For mobile device MD k Safety requirements, SR k ∈{SR1,...,SR r}, used to characterize mobile device MD k The required offloading reliability becomes more stringent as the SR level increases; DA k Indicates mobile device MD k Required data attributes include packet count, port number, minimum packet arrival time, etc. The MEC server provides priority resources for computational tasks with higher user level (UL) and service level (SL).

[0078] MID k For mobile device MD k A unique resource matching identifier indicating a specific unloading point MEC.s ,s∈S. MBD k It describes the matching behavior on the supplier's side, that is Among them, R k For mobile device MD k Released task unloading rate; D k Size of allocated computing resources; LN k This is the link number that was used during the uninstallation process. Therefore, access space AS k and matching space MS k The mapping matrix is:

[0079]

[0080] In step S103, based on the identifier space mapping model, the energy cost and time cost of the computing task corresponding to the computing request are calculated, and a multi-objective optimization model is constructed.

[0081] In this embodiment of the invention, in order to clarify the specific situation of energy and time consumption in the computing task, the energy and time costs of the user end and the server end in the computing task are calculated separately, so as to clarify the optimization goal of the computing task.

[0082] In step S104, the optimal allocation strategy for computational tasks is determined based on the identifier space mapping model and the multi-objective optimization model.

[0083] In this embodiment of the invention, in order to solve the multi-objective optimization problem of energy cost and time cost, the resource orchestration strategy is adjusted and optimized through an optimization model to obtain the resource allocation strategy with the minimum energy cost and time cost, so as to support efficient and reliable resource allocation.

[0084] Through the above embodiments, by constructing a system model that maps the identifier space, the computing tasks of the edge network system are mapped to the identifier space, resulting in an optimization model for the energy and time costs of the edge network system. This optimization model is then used to adjust and optimize the resource orchestration strategy, yielding a resource allocation strategy that minimizes both energy and time costs, thus supporting efficient and reliable resource allocation. By considering the multi-dimensional attribute values ​​of terminals and services, the reliability and network performance of the edge network system are simultaneously guaranteed. Furthermore, by matching the relationship between supply and demand, the reliability, energy consumption, and power consumption of the edge network system are jointly optimized.

[0085] In one embodiment, based on an identifier space mapping model, the energy cost and time cost of the computing task corresponding to the computing request are calculated to construct a multi-objective optimization model, including: calculating the energy cost and time cost of the user-side computing task in the computing task based on an access mapping matrix to obtain a user-side energy cost and time cost optimization problem; calculating the energy cost and time cost of the server-side computing task in the computing task based on a matching mapping matrix to obtain a server-side energy cost and time cost optimization problem; and constructing a multi-objective optimization model based on the user-side energy cost and time cost optimization problem and the server-side energy cost and time cost optimization problem.

[0086] In one example, in an edge network system, each mobile device MD k For each k∈{1,...,K}, there is a computational task to be started. This task can contain several subtasks, which can be executed locally or partially offloaded to the MEC server. System time can be divided into several blocks, each block's time not exceeding the convergence time of the MEC server application or the channel coherence time. To avoid mobile device MD... k Interference between each mobile device MD k The total time for the computation task is set as T = {t1,...,t} 2m}. In t m Before starting, mobile device MD k Continuously sending computation requests to the MEC server; in t m Once completed, the MEC server returns the calculation results to the mobile device MD in sequence. k Since the cost during the return process is very small, the energy and time consumption during the feedback process can be ignored. (Mobile device MD) k The total energy consumption and time cost of the computing task only involve the mobile device MD. k The local execution process, the computation unloading process, and the MEC server computation process.

[0087] For local execution processes, mobile device MD k Local execution time for:

[0088]

[0089] Among them, l k For mobile device MD k The size of the locally executed task, l k ≥0, c k For mobile device MD k The number of CPU cycles required to execute a computational task locally, expressed in cycles per bit; τ kFor mobile device MD k Delay constraint; f k For mobile device MD k The CPU frequency, measured in Hz. Among them, the mobile device MD... k CPU frequency f k It cannot exceed the maximum frequency that the CPU can withstand, that is:

[0090]

[0091] Due to time constraints, mobile device MD k The total time period T of the computation task cannot exceed the mobile device MD. k Delay constraint τ k Latency limit, mobile device MD k Local execution time It cannot exceed the time constraint of the MEC system, that is:

[0092]

[0093] Based on the above constraints, the MD of the mobile device can be derived. k Locally executed tasks are limited to:

[0094]

[0095] Among them, l k For mobile device MD k The size of the locally executed task, l k ≥0; G k As the total number of tasks, G k ≥0.

[0096] Mobile Device MD k Energy cost of local execution for:

[0097]

[0098] Where, ε k For mobile device MD k Effective capacitance coefficient, ε k >0.

[0099] For the computational unloading process, once the mobile device MD... k A compute task request was initiated by the gateway GW. r The appropriate offloading locations and rates will be matched based on information level, resource and security requirements, thereby achieving an effective match between demand and supply.

[0100] cf k For mobile device MDk The conditions for initiation, Among them, mobile device MD k User-level UL k Mobile device MD k Requested service type SL k and mobile device MD k Safety Requirements SR k The higher, cf k The larger the value, the more variable the weights of the three parts, α1, α2, and α3.

[0101] W represents the mobile device MD. k The channel selection matrix, W = {w k ,k∈K}, where w k For mobile device MD k The number of channels occupied, via w k =γ k cf k γ was calculated to be... k It is a direct proportionality coefficient.

[0102] P represents the mobile device MD. k The transmission power matrix, P = {p k p, k∈K} k This represents the transmission power. Therefore, the transmission rate R is... MD,k It can be represented as:

[0103]

[0104] Where B is the channel bandwidth in the edge network system, L is the total number of channels in the edge network system, and MD is the mobile device. k Number of channels occupied ∑ k∈K w k ≤L;g k Let σ be the channel gain, and σ be the Gaussian white noise variable. Therefore, the time cost of offloading, i.e., the transmission delay, is calculated. for:

[0105]

[0106] Among them, CQ k For mobile device MD k Some computing tasks are offloaded to the MEC server, namely And G k -l k ≥0.

[0107] Due to reliable multi-path uninstallation, MD k Safety requirements SR kThe higher the value, the more likely the link number LN is to be used. k The larger the value, the greater the impact on transmission power consumption. Setting LN... k =β k SR k , where β k It is a direct proportionality coefficient. Therefore, the mobile device MD k Calculate the energy cost of unloading for:

[0108]

[0109] For the computation process on the MEC server, assume the MEC server's computing capacity is large enough, and the upper limit of the MEC server cluster's computing capacity is D. Therefore, the MEC server computation process time... for:

[0110]

[0111] D k =μ k cf k

[0112]

[0113] Among them, D k The computing resources provided for the MEC server, μ k This refers to the unit computing power of the MEC server.

[0114] MEC server computing process and mobile device MD k The demand is directly proportional to the demand, as shown below:

[0115]

[0116] Among them, v k v is the energy consumption ratio coefficient. k =n k cf k n k This is the unit loss coefficient.

[0117] In one example, constructing a multi-objective optimization model includes: posing a multi-objective optimization problem concerning time and total energy:

[0118] The total energy and time consumption of an edge network system includes:

[0119]

[0120] In an edge network system, assuming the computing power of the MEC server is unlimited, the size of the computing tasks of all mobile device MDs, the CPU cycles required for local computing on the mobile device MDs, and the energy cost per unit on the MEC server are all known. Therefore, the total time cost optimization problem of the edge network system can be modeled as follows:

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131] Where, θ k For mobile device MD k Weight in time cost. C1 represents the mobile device's MD (Mean) value. k The upper and lower bounds of the local CPU frequency; C2 represents the mobile device MD. k The constraints between local processing time, unloading time, and latency limits; C3 represents the device MD k The range of locally computed bits; C4 represents the channel constraint; C5 represents the MD. k The total time cannot exceed the latency limit; C6 ensures that the total computing resources allocated to the MEC server cannot exceed the total capacity; C7 indicates the total offload latency limit; C8 indicates the offload energy limit.

[0132] Similarly, the total energy cost optimization problem for edge network systems can be derived as follows:

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143] Where, ψ k Indicates mobile device MD k The weight of the edge network system in the total energy cost, therefore, the total energy cost optimization problem can be simplified to:

[0144]

[0145]

[0146]

[0147]

[0148]

[0149]

[0150]

[0151]

[0152] Based on the above optimization problem, energy cost and time cost can be expressed as a multi-objective optimization problem, enabling edge network systems to achieve reliable and effective resource utilization.

[0153] Therefore, the multi-objective optimization model is expressed as:

[0154]

[0155]

[0156] stC1-C8

[0157] Furthermore, the optimization problem is transformed into a unified format, namely:

[0158]

[0159] stC1-C8

[0160] Where, ζ k Indicates mobile device MD k The weights in terms of energy and time consumption, ζ k ==l(θ) k +ψ k According to the task requester's mobile device MD. k Based on the preferences and weighted proportions, feasible solutions for optimizing energy and time costs are obtained. When ζ k The larger the size, the higher the demand for mobile devices MD in the access space AS. k The higher the priority, the more resources are occupied and consumed.

[0161] In one embodiment, the optimal allocation strategy for computing tasks is determined based on an identifier space mapping model and a multi-objective optimization model, including: determining the resource supply and demand relationship between the server and the user based on an access mapping matrix; determining the optimal matching position based on the resource supply and demand relationship between the server and the user; constructing a matching strategy optimization model based on the multi-objective optimization model and the optimal matching position; and determining the optimal allocation strategy for computing tasks based on the matching strategy optimization model.

[0162] In this embodiment of the invention, determining the resource supply and demand relationship between the server and the user based on the access mapping matrix includes: obtaining a supply and demand relationship model; inputting the access mapping matrix into the supply and demand relationship model to obtain the resource supply and demand relationship between the server and the user. The supply and demand relationship model is a pre-trained deep neural network matching model.

[0163] In one example, the process of determining the optimal matching location may include: using the access space AS information as input to the matching model and the unloading location MID as output of the matching model, through a pre-trained DNN matching model.

[0164] The DNN matching model comprises a separate CRNN module, a model using both CNN and LSTM. The CNN acts as the encoder, converting accessed spatial information into a multi-dimensional matrix pattern and extracting spatial information, while the LSTM decoder parses the matrix information into multiple one-dimensional vectors. The encoding process consists of convolution, activation, pooling, and fully connected operations. During decoding, three bidirectional LSTM modules are developed, reflected in the forget gate, input gate, internal state gate, and output gate, respectively. Specifically, the first two LSTM modules have 256 noisy activation units, while the third layer has 218 tanh activation units. The final part of the DNN matching model consists of several s-dimensional fully connected layers. To achieve more accurate predictions, a controlled variable approach is used to find the optimal result, considering factors such as the number of features, the proportion of the training set, the number of input data points, and the number of epochs. Furthermore, to avoid overfitting, a dropout technique with a probability of 0.2 is added to all LSTM layers.

[0165] In the training matching samples, the optimal offloading point for computational tasks is obtained by mapping specific service requests to specific MEC servers. Assuming each MEC server has a specific service type, sample collection is achieved by mapping predicted service types to corresponding servers. The collected samples are based on 24-hour full-duplex traces of a single location. This site is a research facility's mainframe, with approximately 1000 devices connected to the internet via a full-duplex gigabit Ethernet link. Information originates from the TCP transport protocol, and specific details in the features are obtained by individually calculating packet header information. The input data is preprocessed to improve its training performance, namely, missing information imputation, normalization, and one-hot encoding. This preprocessed data becomes the input to the encoder-decoder model. The model then extracts the offloading patterns of computational tasks through the encoding and decoding processes and continuously optimizes the weights and biases using an adaptive moment estimation algorithm.

[0166] To represent the degree of matching between resource supply and demand, we introduce a matching index, namely the matching error ME, and convert the model output into ME form. ME, as the model's loss function, consists of task execution efficiency (TEU) and task cost efficiency (TCU), and can be expressed as follows:

[0167] ME=ω TEU TEU+ω TCU TCU

[0168] ω TCU +ω TCU =1

[0169] Where, ω TCU ω represents the weighting coefficient for performance effectiveness. TCUThe weighting coefficients represent cost-utility. The expected utility function theory is introduced to calculate the task execution efficiency (TEU) and task cost efficiency (TCU) during the task unloading process. The task execution efficiency (TEU) is calculated based on the execution capacity matching of the MEC server and task requirements, as shown below:

[0170]

[0171] sl≥sl *

[0172] Where sl represents the computing resources that the MEC server cluster can provide for computing requests, sl * This represents the minimum computing resources consumed by the total offloading tasks within the same system cycle. sl and sl * It can be calculated using the expected utility function:

[0173]

[0174]

[0175] in, MD k The computational tasks are offloaded to the MEC server. i The probability, Indicates server MEC i For MD k The maximum resource provided, r k MD k Minimum resources consumed on the MEC server side.

[0176] Task Cost Efficiency (TCU) measures the cost balance between suppliers and requesters, and can be expressed as:

[0177]

[0178] c max ≥c*

[0179] Among them, c max This indicates the maximum cost that an MEC server cluster can bear or the maximum return that MD can provide. * This represents the total cost required for the MEC service cluster to execute the task volume. Similarly, c max and c * It can be calculated using the expected utility function:

[0180]

[0181] in, For MEC i The server executes the mobile device MD. kThe maximum tolerable cost of the total task, p k For mobile device MD k The payment, This represents the actual calculated cost.

[0182] When the matching error ME approaches 0, the model has reached equilibrium, and the task unloading process is close to optimal. To achieve the objective while satisfying the constraints, a weighted sum of the objective function, i.e., the overall matching error ME, needs to be considered. Through continuous iteration of the model, ME is continuously optimized, i.e., minME = ω. YEU TEU+ω TCU TCU. Training ends when ME≤ε, resulting in multiple one-dimensional vectors. The label corresponding to the highest probability value is selected from these vectors as the prediction result for multiple computational requirements. The training phase can be completed offline.

[0183] In this embodiment of the invention, determining the optimal allocation strategy for computing tasks based on the matching strategy optimization model includes: determining an initial matching strategy based on a multi-objective optimization model; and adjusting the initial matching strategy based on the matching strategy optimization model to minimize the energy cost and time cost of the system, thereby obtaining the optimal allocation strategy for computing tasks.

[0184] In one example, the process of determining the optimal allocation strategy for computational tasks may include: Since quantum genetic algorithms (GGAs) do not restrict the continuity and direction of the optimization problem, they are used to find the optimal resource allocation. The optimal supplier-side matching strategy (MBD) configuration is obtained by optimizing the transmission power allocation p, the local computational capacity allocation f, the channel resource allocation ω, and the computational resource allocation D in the server MECs, while simultaneously satisfying the minimum energy consumption of the MBD. A competition-based approach is used to select the next generation of individuals, similar to natural selection, making it easier for individuals with higher habitability value to survive to the next generation, i.e., individuals with lower system consumption are more likely to survive. The optimal solution for resource allocation can be obtained before the iteration limit reaches or falls below a certain threshold.

[0185] 1. Chromosome Construction and Initialization: Since genetic algorithms cannot directly handle the spatial solution parameters of a problem, they need to be converted into chromosomes composed of genes according to a certain structure. In quantum genetic algorithms, chromosomes correspond to solutions to the optimization problem. Each chromosome encodes a feasible solution to a resource allocation problem, and the chromosome length L... c Number of tasks. Each gene molecule represents a candidate deployment corresponding to one task. The chromosome is an ordered sequence of the L gene. Where c i ∈[1,n],1≤i≤L cThis represents the task allocation scheme, with n candidate allocation schemes. The p, f, ω, and D of the optimization problem constitute the chromosome information of an individual. The population size is set to g. For a single C... j It can be represented as the following matrix:

[0186]

[0187]

[0188] in During initialization, we randomly generate an initial population, which is the first set of chromosomes in the search process. We set the initial candidate chromosome set as... To ensure all chromosomes reach a uniform superposition of 8, a Hadamard gate is applied to each individual gene on each chromosome. The j-th chromosome is initialized as follows:

[0189]

[0190] Where |φ> j =C j Each chromosome is simply an l-bit binary string.

[0191] 2. Fitness Function: The fitness function is an indicator of chromosome quality. Generally, this algorithm selects the optimal individual based on the value of the fitness function. Therefore, individuals with poor health are excluded. For a single j (1≤j≤g) in the t-th iteration, the fitness function is defined as:

[0192]

[0193] Where f is the objective function. The value M in the fitness function is to ensure that the fitness is non-negative, facilitating the calculation of evolutionary probabilities. Habitability is defined as an evaluation criterion for assessing the quality of solutions generated during genetic iteration. It can be expressed as:

[0194]

[0195] Clearly, higher habitability corresponds to a higher fitness value, which indirectly corresponds to a smaller objective function, thus resulting in a better individual. Individuals with high habitability have a greater chance of reproducing in the next generation. Otherwise, the individual represents a less viable solution. During the iteration process, the fitness function is a necessary condition to guarantee the feasibility of candidate objects. Once an abnormal individual is encountered during the search process, the algorithm will restart the search process.

[0196] 3. Selection, crossover, and mutation: These three operations are the basic operators for generating subpopulations. The selection process reflects the natural law of survival of the fittest, the crossover process embodies the idea of ​​sexual reproduction, and the mutation operation embodies gene mutation in the evolutionary process.

[0197] In the selection process, we use a competition method to select contemporary individuals, i.e., the parental group for the next generation. Each time, g / 2 individuals are randomly selected from the population. Then, the two individuals with the highest survival rates are selected from the competition. This process is repeated until the number of parental candidate species reaches the population size g. After selecting the candidates, a measurement is used to map G(t) to P(t). Based on the measurement results... And the current best result b j The chromosome rotation is updated according to the update rules:

[0198]

[0199] In the formula For the j-th gene on the i-th chromosome, b j Let be the j-th gene in the considered optimal chromosome, and δθ be the increment or decrement of the rotation. The crossover process affects the convergence speed of the algorithm. In each round, two parent individuals are selected for chromosome crossover or crossover with probability p. c Recombination is performed, producing two offspring individuals to participate in the next iteration. Superior genes from parents can be inherited, allowing the solution to the problem to evolve in a direction that reduces the value of the objective function. Mutation refers to the generation of new offspring through random gene mutations. The mutation process of quantum chromosomes can be achieved by reversing the amplitudes of α and β through an X-gate. This process avoids the genetic algorithm getting trapped in local optima and instead propels it towards the global optimum.

[0200] Finally, repeat the above process until the iteration limit is reached or the habitability reaches the threshold. This will ultimately yield the optimal resource allocation and deployment.

[0201] Based on the same inventive concept, the present invention also provides an edge network resource allocation device.

[0202] Figure 3 This is a structural block diagram of an edge network resource allocation device according to an exemplary embodiment. Figure 3 As shown, the edge network resource allocation device includes an acquisition unit 301, a mapping unit 302, a calculation unit 303, and a determination unit 304.

[0203] Acquisition unit 301 is used to acquire the user's calculation request;

[0204] The mapping unit 302 is used to map the computation request to the identifier space based on the computation request to obtain the identifier space mapping model. The identifier space mapping model is used to represent the mapping of the relationship between the user and the server.

[0205] The computing unit 303 is used to calculate the energy cost and time cost of the computing task corresponding to the computing request based on the identifier space mapping model, and to construct a multi-objective optimization model.

[0206] The determination unit 304 is used to determine the optimal allocation strategy for computing tasks based on the identifier space mapping model and the multi-objective optimization model.

[0207] In one embodiment, the identifier space includes an access space and a matching space; the mapping unit 302 includes: a first determining unit, configured to determine a computing task based on a computing request, the computing task including a computing unloading task and a resource allocation task; an access mapping unit, configured to map the computing unloading task to the access space to obtain an access mapping matrix; a matching mapping unit, configured to map the resource allocation task to the matching space to obtain a matching mapping matrix; and a space mapping unit, configured to match the mapping matrix based on the access mapping matrix to obtain an identifier space mapping model.

[0208] In another embodiment, the computing unit 303 includes: a user-side computing unit, configured to calculate the energy cost and time cost of the user-side computing task in the computing task based on the access mapping matrix, to obtain a user-side energy cost and time cost optimization problem; a server-side computing unit, configured to calculate the energy cost and time cost of the server-side computing task in the computing task based on the matching mapping matrix, to obtain a server-side energy cost and time cost optimization problem; and a first construction unit, configured to construct a multi-objective optimization model based on the user-side energy cost and time cost optimization problem and the server-side energy cost and time cost optimization problem.

[0209] In another embodiment, the determining unit 304 includes: a supply and demand matching unit, used to determine the resource supply and demand relationship between the server and the user based on the access mapping matrix; a location matching unit, used to determine the optimal matching location based on the resource supply and demand relationship between the server and the user; a second construction unit, used to construct a matching strategy optimization model based on a multi-objective optimization model and the optimal matching location; and a strategy determining unit, used to determine the optimal allocation strategy for computing tasks based on the matching strategy optimization model.

[0210] In another embodiment, the supply and demand matching unit includes: a first acquisition unit for acquiring a supply and demand relationship model; and an input unit for inputting an access mapping matrix into the supply and demand relationship model to obtain the resource supply and demand relationship between the server and the user.

[0211] In another embodiment, the supply and demand relationship model is a pre-trained deep neural network matching model.

[0212] In another embodiment, the strategy determination unit includes: a first determination unit, configured to determine an initial matching strategy based on a multi-objective optimization model; and an adjustment unit, configured to adjust the initial matching strategy based on the matching strategy optimization model to minimize the energy cost and time cost of the system, thereby obtaining the optimal allocation strategy for the computing task.

[0213] The specific limitations and beneficial effects of the aforementioned edge network resource allocation device can be found in the limitations of the edge network resource allocation method described above, and will not be repeated here. Each of the above modules can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0214] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an exemplary embodiment. For example... Figure 4 As shown, the device includes one or more processors 410 and a memory 420, which includes persistent memory, volatile memory, and a hard disk. Figure 4 Taking a processor 410 as an example, the device may also include an input device 430 and an output device 440.

[0215] The processor 410, memory 420, input device 430, and output device 440 can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0216] Processor 410 can be a Central Processing Unit (CPU). Processor 410 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof. The general-purpose processor can be a microprocessor or any conventional processor.

[0217] The memory 420, as a non-transitory computer-readable storage medium, includes persistent memory, volatile memory, and a hard disk. It can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the edge network resource allocation method in this embodiment. The processor 410 executes various server functions and data processing by running the non-transitory software programs, instructions, and modules stored in the memory 420, thereby implementing any of the aforementioned edge network resource allocation methods.

[0218] The memory 420 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data that is needed and required. Furthermore, the memory 420 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 420 may optionally include memory remotely located relative to the processor 410, and these remote memories can be connected to the data processing device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0219] Input device 430 can receive input numerical or character information, and generate key signal inputs related to user settings and function control. Output device 440 may include display devices such as a display screen.

[0220] One or more modules are stored in memory 420, and when executed by one or more processors 410, they perform actions such as... Figures 1-2 The method shown.

[0221] The above-described product can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in [reference 1]. Figures 1-2 The relevant descriptions in the illustrated embodiments.

[0222] This invention also provides a non-transitory computer storage medium storing computer-executable instructions that can execute the authentication method in any of the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0223] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for allocating resources in an edge network, characterized in that, The method includes: Obtain the user's computation request; Based on the computation request, the computation request is mapped to the identifier space to obtain the identifier space mapping model, which is used to represent the mapping of the relationship between the user and the server; Based on the identifier space mapping model, the energy cost and time cost of the computing task corresponding to the computing request are calculated, and a multi-objective optimization model is constructed. Based on the identifier space mapping model and the multi-objective optimization model, the optimal allocation strategy for the computing tasks is determined; The multi-objective optimization model is as follows: Where k represents the k-th mobile device K represents the total number of mobile devices. For mobile devices Weight in time cost Indicates mobile device Local execution time, Indicates mobile device Transmission delay, This indicates the time taken for the MEC server to perform the computation. Indicates mobile device Weight in total energy cost Indicates mobile device Energy costs for local execution Indicates mobile device Calculate the energy cost of unloading. , This is the energy consumption ratio coefficient. , The unit loss coefficient, For mobile devices The conditions for initiation, ,and , Indicates mobile device CPU frequency, Indicates mobile device The size of the task executed locally. Indicates mobile device The number of CPU cycles required to execute a computational task locally. This indicates the maximum frequency that the CPU can handle, and T represents the mobile device. The total time cycle of the computation task. Indicates mobile device Delay constraints, Indicates mobile device Total number of tasks Indicates mobile device The number of channels occupied This represents the total number of channels in the edge network system. This represents the total computing resources allocated to the MEC server, where D represents the total capacity. This indicates that energy is being unloaded.

2. The method according to claim 1, characterized in that, The identifier space includes an access space and a matching space; The step of mapping the computation request to the identifier space mapping model based on the computation request includes: Based on the computing request, a computing task is determined, which includes a computing unloading task and a resource allocation task. The computational offloading task is mapped to the access space to obtain the access mapping matrix; The resource allocation task is mapped to the matching space to obtain the matching mapping matrix; The identifier space mapping model is obtained by matching the access mapping matrix with the mapping matrix.

3. The method according to claim 2, characterized in that, The step of calculating the energy cost and time cost of the computing task corresponding to the computing request based on the identifier space mapping model, and constructing a multi-objective optimization model, includes: Based on the access mapping matrix, the energy cost and time cost of the user-end computing task in the computing task are calculated to obtain the user-end energy cost and time cost optimization problem. Based on the matching mapping matrix, the energy cost and time cost of the server-side computing task in the computing task are calculated to obtain the server-side energy cost and time cost optimization problem. Based on the optimization problems of energy cost and time cost on the user side and the optimization problems of energy cost and time cost on the server side, a multi-objective optimization model is constructed.

4. The method according to claim 3, characterized in that, The step of determining the optimal allocation strategy for the computational task based on the identifier space mapping model and the multi-objective optimization model includes: Based on the access mapping matrix, the resource supply and demand relationship between the server and the user is determined; Based on the resource supply and demand relationship between the server and the user, the optimal matching position is determined; Based on the multi-objective optimization model and the optimal matching position, a matching strategy optimization model is constructed; Based on the matching strategy optimization model, the optimal allocation strategy for the computing task is determined.

5. The method according to claim 4, characterized in that, Determining the resource supply and demand relationship between the server and the user based on the access mapping matrix includes: Obtain a supply and demand relationship model; By inputting the access mapping matrix into the supply and demand relationship model, the resource supply and demand relationship between the server and the user is obtained.

6. The method according to claim 5, characterized in that, The supply and demand relationship model is a pre-trained deep neural network matching model.

7. The method according to claim 4, characterized in that, The step of determining the optimal allocation strategy for the computational task based on the matching strategy optimization model includes: Based on the multi-objective optimization model, determine the initial matching strategy; Based on the matching strategy optimization model, the initial matching strategy is adjusted to minimize the energy and time costs of the system, thereby obtaining the optimal allocation strategy for the computing task.

8. An edge network resource allocation device, characterized in that, The device includes: The acquisition unit is used to acquire the user's computing request; A mapping unit is used to map the computation request to an identifier space based on the computation request, thereby obtaining an identifier space mapping model, which is used to represent the mapping of the relationship between the user and the server. The computing unit is used to calculate the energy cost and time cost of the computing task corresponding to the computing request based on the identifier space mapping model, and to construct a multi-objective optimization model. The determining unit is used to determine the optimal allocation strategy for the computing task based on the identifier space mapping model and the multi-objective optimization model. The multi-objective optimization model is as follows: Where k represents the k-th mobile device K represents the total number of mobile devices. For mobile devices Weight in time cost Indicates mobile device Local execution time, Indicates mobile device Transmission delay, This indicates the time taken for the MEC server to perform the computation. Indicates mobile device Weight in total energy cost Indicates mobile device Energy costs for local execution Indicates mobile device Calculate the energy cost of unloading. , This is the energy consumption ratio coefficient. , The unit loss coefficient, For mobile devices The conditions for initiation, ,and , Indicates mobile device CPU frequency, Indicates mobile device The size of the task executed locally. Indicates mobile device The number of CPU cycles required to execute a computational task locally. This indicates the maximum frequency that the CPU can handle, and T represents the mobile device. The total time cycle of the computation task. Indicates mobile device Delay constraints, Indicates mobile device Total number of tasks Indicates mobile device The number of channels occupied This represents the total number of channels in the edge network system. This represents the total computing resources allocated to the MEC server, where D represents the total capacity. This indicates that energy is being unloaded.

9. A computer device, characterized in that, The device includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the edge network resource allocation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the edge network resource allocation method according to any one of claims 1-7.

Images