Resource allocation method and resource allocation apparatus

Abstract

The application relates to a resource allocation method and a resource allocation device. The method comprises the following steps: receiving at least one computing task request sent by a power internet of things device; distributing each computing task request to a low-delay network slice and a high-throughput network slice; calculating a first task demand parameter of a first computing task request included in the low-delay network slice; calculating a second task demand parameter of a second computing task request included in the high-throughput network slice; and determining the computing power resource allocated to each computing task request according to the first task demand parameter and the second task demand parameter. The method can improve the utilization rate of the computing power resource.

Description

Technical Field

[0001] This application relates to the field of Internet of Things (IoT) technology, and in particular to a resource allocation method and a resource allocation device. Background Technology

[0002] With the continuous development of cloud computing and edge computing technologies, more and more large-scale network architectures are beginning to process data by combining edge computing with cloud computing. For example, the Internet of Things (IoT) for power systems uses edge computing to push cloud computing capabilities down to the edge, reducing the data processing pressure on cloud computing centers.

[0003] However, the Internet of Things for power needs to handle a large number of task requests, and different tasks require different network resources. How to improve resource utilization has become a concern for all parties. Summary of the Invention

[0004] Therefore, it is necessary to provide a resource allocation method and device that can improve resource utilization in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a resource allocation method applied to an edge computing server, comprising: receiving at least one computing task request sent by a power Internet of Things (IoT) device; allocating each computing task request to a low-latency network slice and a high-throughput network slice; calculating a first task requirement parameter for a first computing task request included in the low-latency network slice, and calculating a second task requirement parameter for a second computing task request included in the high-throughput network slice; and determining the computing power resources allocated to each computing task request based on the first task requirement parameter and the second task requirement parameter.

[0006] In one embodiment, determining the computing resources allocated to each computing task request based on the first task requirement parameter and the second task requirement parameter includes: weighting and summing the first task requirement parameter and the second task requirement parameter to obtain a total requirement parameter; subtracting the system cost parameter from the total requirement parameter to obtain a system utility function, where the system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each computing task request; and solving for the maximum value of the system utility function to obtain the computing resources allocated to each computing task request.

[0007] In one embodiment, the system utility function is maximized to obtain the computing resources allocated to each computing task request. This includes: iteratively solving the neural network and gradient network based on the system utility function, and obtaining the computing resources allocated to each computing task request when the neural network and gradient network converge. Specifically, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network, and the system utility function and the computing device information are input into the gradient network to obtain the computing resources corresponding to each computing task request.

[0008] In one embodiment, when the neural network fails to converge, the method further includes: for each first computing task request in each computing task request, using the first task requirement parameter corresponding to the first computing task request as a first state space, the computing device information corresponding to the first computing task request as a first action space, and the system utility function corresponding to the first computing task request as a first reward function; determining a first loss value based on the first state space, the first action space, and the first reward function, and adjusting the parameters of the neural network based on the first loss value.

[0009] In one embodiment, if the gradient network fails to converge, the method further includes: for each second computational task request in each computational task request, using the second task requirement parameters corresponding to the second computational task request as a second state space, the computing power resources corresponding to the second computational task request as a second action space, and the system utility function corresponding to the second computational task request as a second reward function; determining a second loss value based on the second state space, the second action space, and the second reward function; and adjusting the parameters of the gradient network based on the second loss value.

[0010] In one embodiment, calculating the first task requirement parameters for a first computing task request included in a low-latency network slice includes: for each first computing task request, determining a first latency and a second latency, where the first latency represents the processing latency between the first computing task request and the edge computing server, and the second latency represents the processing latency between the first computing task request and the cloud computing center; determining a first index based on a comparison between the first latency and a reference latency, where the first index represents the service quality assurance satisfaction level between the first computing task and the edge computing server; determining a second index based on a comparison between the second latency and the reference latency, where the second index represents the service quality assurance satisfaction level between the first computing task and the cloud computing center; and determining the first task requirement parameters based on the first index and the second index.

[0011] In one embodiment, for each first computing task request, determining a first delay and a second delay of the first computing task request includes: for each first computing task request, determining a first transmission delay and a first edge computing delay corresponding to the first computing task request, and adding the first transmission delay and the first edge computing delay to obtain a first delay, wherein the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed on the edge computing server; determining a second transmission delay and a third transmission delay corresponding to the first computing task request, and adding the second delay and the third delay to obtain a second delay, wherein the second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third delay represents the wired transmission delay when the first computing task request is processed in the cloud computing center.

[0012] In one embodiment, calculating the second task requirement parameters for the second computing task requests included in the high-throughput network slice includes: for each second computing task request, determining a third latency, a fourth latency, and bandwidth requirement parameters for the second computing task request, wherein the third latency characterizes the processing latency corresponding to the second computing task request and the edge computing server, and the fourth latency characterizes the processing latency corresponding to the second computing task request and the cloud computing center; based on the comparison result between the third latency and the reference latency, determining a third index, wherein the third index characterizes the service quality assurance satisfaction when the second computing task is computed on the edge server; based on the comparison result between the fourth latency and the reference latency, determining a fourth index, wherein the fourth index characterizes the service quality assurance satisfaction when the second computing task request is computed in the cloud computing center; and calculating the weighted average sum of the third index, the fourth index, and the bandwidth requirement parameters to obtain the second task requirement parameters.

[0013] In one embodiment, for each second computing task request, a third delay, a fourth delay, and bandwidth requirement parameters for the second computing task request are determined, including: for each second computing task request, determining a fourth transmission delay and a second edge computing delay corresponding to the second computing task request, and adding the fourth transmission delay and the second edge computing delay to obtain a third delay, wherein the fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server; determining a fifth transmission delay and a sixth transmission delay corresponding to the second computing task request, and adding the fifth delay and the sixth delay to obtain a fourth delay, wherein the fifth delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth delay represents the wired transmission delay between the second computing task request and the cloud computing center; and determining a lower bound and a pre-allocated bandwidth corresponding to the second computing task request, and defining the lower bound and the pre-allocated bandwidth as bandwidth requirement parameters.

[0014] Secondly, this application also provides a resource allocation apparatus, comprising: a task request allocation module, configured to receive at least one computing task request sent by a power Internet of Things (IoT) device, and allocate each computing task request to a low-latency network slice and a high-throughput network slice; a demand parameter calculation module, configured to calculate a first task demand parameter for a first computing task request included in a low-latency network slice, and calculate a second task demand parameter for a second computing task request included in a high-throughput network slice; and a computing power resource determination module, configured to determine the computing power resources allocated to each computing task request based on the first task demand parameter and the second task demand parameter.

[0015] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect.

[0016] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0017] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0018] The aforementioned resource allocation method and apparatus involve an edge computing server first receiving at least one computing task request from a power IoT device, allocating each computing task request to a low-latency network slice and a high-throughput network slice. The edge computing server calculates a first task requirement parameter for the first computing task request included in the low-latency network slice based on the task requirements of the computing task requests in the low-latency network slice, and calculates a second task requirement parameter for the second computing task request included in the high-throughput network slice based on the task requirements of the computing task requests in the high-throughput network slice. The computing power resources allocated to each computing task request are determined based on the first and second task requirement parameters. This accurate allocation method avoids the situation where some tasks receive additional computing power resources while other tasks suffer from insufficient computing power resources, thereby improving the utilization rate of computing power resources. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a diagram illustrating the application environment of a resource allocation method in one embodiment.

[0021] Figure 2 This is a flowchart illustrating a resource allocation method in one embodiment;

[0022] Figure 3 This is a flowchart illustrating the steps for determining the first task requirement parameters in one embodiment;

[0023] Figure 4 This is a flowchart illustrating step 301 in one embodiment;

[0024] Figure 5 This is a flowchart illustrating the steps for determining the second task requirement parameters in one embodiment;

[0025] Figure 6 This is a flowchart illustrating step 501 in one embodiment;

[0026] Figure 7 This is a flowchart illustrating step 203 in one embodiment;

[0027] Figure 8 This is a flowchart illustrating step 801 in one embodiment;

[0028] Figure 9 This is a flowchart illustrating the parameter tuning steps of a neural network in one embodiment;

[0029] Figure 10 This is a flowchart illustrating the parameter tuning steps of a gradient network in one embodiment.

[0030] Figure 11 This is a flowchart illustrating the resource allocation method in another embodiment;

[0031] Figure 12 This is a structural block diagram of a resource allocation device in one embodiment;

[0032] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0034] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0035] The resource allocation method provided in this application embodiment can be applied to, for example, Figure 1 The application environment shown includes at least a power Internet of Things (IoT) terminal set 101, an edge computing server 102, a base station 103, and a cloud computing center 104.

[0036] The terminal set 101 is used to send at least one computing task request to the edge computing server 102 via the base station 103. The terminal set 101 can be various types of power terminals, including various power equipment, power drones, power IoT devices, etc.

[0037] Edge computing server 102 is used to acquire at least one computing task request sent by terminal set 101 via base station 103, allocate each computing task request to low-latency network slices and high-throughput network slices, calculate a first task requirement parameter for a first computing task request included in the low-latency network slice, and calculate a second task requirement parameter for a second computing task request included in the high-throughput network slice, and determine the computing resources allocated to each computing task request based on the first and second task requirement parameters. Edge computing server 102 can also be used to process tasks sent by terminal set 101 based on the allocation result of computing resources.

[0038] Base station 103 can be used for network communication between terminal set 101, edge computing server 102, and cloud computing center 104. Cloud computing center 104 can be used to process tasks sent by terminal set 101 according to the allocation results of computing resources.

[0039] In real-world scenarios, the power sector primarily comprises five major business segments: power generation, transmission, transformation, distribution, and consumption. Each segment involves distinct power business scenarios. Integrating power with technologies such as the Internet of Things (IoT), 5G, and artificial intelligence (AI) to form the power IoT is a crucial means for the power grid to move towards digitalization and intelligence. The power IoT deploys a large number of IoT terminals, which collect massive amounts of power operation data. To extract useful information from this vast amount of power data to assist in power production and management, it is necessary to utilize AI algorithms and other technologies to perform rapid and efficient calculations on this data.

[0040] Cloud computing has achieved significant success in high-performance computing. However, data transmission from power IoT terminals to remote cloud computing centers not only experiences considerable latency, failing to meet the low-latency requirements of latency-sensitive power services such as distribution network differential protection and distributed feeder automation, but also exposes power data to the public internet, posing a security risk of data leakage. Edge computing, by deploying edge servers at the network edge and bringing cloud computing capabilities closer to the data source, can reduce network transmission latency and mitigate data security risks. However, compared to cloud computing, edge servers have relatively limited computing power and cannot simultaneously handle diverse business needs. Therefore, it is necessary to coordinate the computing and network resources of cloud computing centers and edge servers, and through cloud-edge-network integration and collaborative edge computing, meet the diverse quality of service requirements of power IoT applications.

[0041] On the other hand, according to power safety requirements, power 5G services can be divided into a production control zone (including security zone I and security zone II), a management information zone (including security zone III and security zone IV), and an internet zone. Not only do each zone have rich power service scenarios and differentiated service quality requirements, but the production control zone also requires resource isolation from other zones. Network slicing can use virtualization technology to isolate physical networks into independent logical networks, each of which is called a slice, with logical resource isolation between slices. Network slicing is an important means of achieving isolation between the production control zone and other security zones.

[0042] Currently, industry and academia have been exploring the power Internet of Things (IoT) based on 5G network slicing and edge computing. The current approach primarily allocates power network slices at the power security zone level, reducing end-to-end transmission latency by offloading latency-sensitive computing tasks to computing nodes at the virtual network edge within the slice. This method provides isolated and edge computing services for power services in different security zones. Due to the diversity of power services and the dynamic nature of the wireless environment, the communication and computing resource requirements of power services change dynamically. To meet the quality of service (QoS) requirements of power services, industry often achieves QoS assurance by over-provisioning slice resources. Over-provisioning slice resources leads to resource waste and increased operating costs; however, insufficient resource provision cannot meet the low latency, high throughput, and efficient computing service requirements of power IoT applications, thus reducing user experience.

[0043] To address this, this application distinguishes the specific task types of various computing task requests and allocates each computing task request to low-latency network slices and high-throughput network slices. The computing power resources allocated to each computing task request are determined by the task demand parameters in the computing network slice. This avoids the situation where some tasks receive additional computing power resources while other tasks lack sufficient computing power resources, thereby improving the utilization rate of computing power resources.

[0044] In one exemplary embodiment, such as Figure 2 As shown, a resource allocation method is provided, which can be applied to... Figure 1 The following steps, 201 to 203, are used as an example of edge computing servers.

[0045] Step 201: Receive at least one computing task request sent by a power IoT device, and allocate each computing task request to a low-latency network slice and a high-throughput network slice.

[0046] During implementation, the edge computing server receives at least one computing task request from a power IoT device. For each computing task request, it extracts the task information and determines whether the task is a low-latency task based on the task information. If the task is in a low-latency network, it is placed in a low-latency network slice; if the task is not in a low-latency network, it is placed in a high-throughput network slice. The low-latency task can be determined based on the task type. The task information includes at least one of the following: task type, task size, task computation volume, and task quality of service requirements.

[0047] Previously, two end-to-end network slices could be defined: a low-latency network slice and a high-throughput network slice. Correspondingly, low-latency tasks are characterized by high sensitivity to latency, such as tasks for real-time communication services, while high-throughput tasks are characterized by high bandwidth requirements, such as tasks for video services.

[0048] Step 202: Calculate the first task requirement parameters of the first computing task request included in the low-latency network slice, and calculate the second task requirement parameters of the second computing task request included in the high-throughput network slice.

[0049] In this application, the task requirement parameters are used to characterize the resources required by the computing task request, and the task requirement parameters may include at least one of network resource requirement parameters, computing resource requirement parameters, and latency requirement parameters.

[0050] During implementation, the edge computing server can calculate the task requirement parameters for each first computing task request included in the low-latency network slice, and determine the first task requirement parameters based on the task requirement parameters of all first computing task requests. Correspondingly, the edge computing server can calculate the task requirement parameters for each second computing task request included in the high-throughput network slice, and determine the second task requirement parameters based on the task requirement parameters of all second computing task requests.

[0051] Step 203: Determine the computing resources to be allocated to each computing task based on the first task requirement parameters and the second task requirement parameters.

[0052] During implementation, the edge computing server can solve for system utility maximization based on the first and second task requirement parameters to obtain the computing resources allocated to each computing task request. The obtained computing resources allocated to each computing task request minimize the total system cost.

[0053] In this application, system utility can be defined as the difference between the weighted sum of the average quality of service (QoS) satisfaction of tasks in low-latency network slices and the average QoS satisfaction of tasks in high-throughput network slices, and the total system cost; the total system cost is the weighted sum of the total communication resources and total computing resources allocated to tasks in low-latency and high-throughput network slices, and the weighted sum of the average latency timeout penalty for tasks in low-latency network slices; the average QoS satisfaction of tasks in low-latency network slices can be a first requirement parameter, and the average QoS satisfaction of tasks in high-throughput network slices can be a second requirement parameter.

[0054] Furthermore, after obtaining the computing resources requested for each computing task, the computing resources requested for each computing task can be sent to the base station to be sent to the corresponding power IoT terminal; the SDN controller included in the power IoT corresponding to the base station configures the communication and computing resources of low-latency slices and high-throughput slices according to the computing resources requested for each computing task. Among them, the wireless resource configuration method in the communication resources is: the available wireless channel resources are allocated to low-latency slices and high-throughput slices according to the ratio of the sum of wireless channel resources allocated to tasks in low-latency slices to the sum of wireless channel resources allocated to tasks in high-throughput slices in the optimal solution.

[0055] Specifically, the bandwidth resource allocation method for communication resources is as follows: the bandwidth of the path from the base station to the cloud computing center is allocated to low-latency slices and high-throughput slices according to the ratio of the sum of bandwidth allocated to tasks in low-latency slices to the sum of bandwidth allocated to tasks in high-throughput slices in the optimal solution. The computing resource allocation method is as follows: the computing resources of the side servers are allocated to low-latency slices and high-throughput slices according to the ratio of the sum of edge server computing resources allocated to tasks in low-latency slices to the sum of edge server computing resources allocated to tasks in high-throughput slices in the optimal solution.

[0056] Furthermore, the SDN controller converts the resource allocation strategy in the optimal solution into network instructions recognizable by network nodes and computing nodes, and sends them to virtual base stations, routers, edge servers, and cloud computing centers in low-latency and high-throughput slices. Correspondingly, after receiving the computing migration decision and network slice identifier, the terminal set of the power Internet of Things sends the computing task to the network slice indicated by the slice identifier according to the computing migration decision. The network slice forwards the task to the corresponding computing node for processing according to the slice identifier of the task and the resource allocation strategy in the optimal solution, and returns the result to the corresponding terminal set of the power Internet of Things.

[0057] In the above resource allocation method, the edge computing server first receives at least one computing task request sent by a power IoT device, and allocates each computing task request to a low-latency network slice and a high-throughput network slice. The edge computing server calculates the first task requirement parameters of the first computing task request included in the low-latency network slice based on the task requirements of the computing task request in the low-latency network slice, and calculates the second task requirement parameters of the second computing task request included in the high-throughput network slice based on the task requirements of the computing task request in the high-throughput network slice. The computing power resources allocated to each computing task request are determined according to the first task requirement parameters and the second task requirement parameters. By accurately allocating resources, the situation where some tasks receive additional computing power resources while other tasks lack sufficient computing power resources is avoided, thereby improving the utilization rate of computing power resources.

[0058] Based on the above exemplary embodiment, the following provides a resource allocation method in one or more exemplary embodiments, which is applied to... Figure 1 Taking edge computing servers as an example, the following content will be used to illustrate this.

[0059] In determining the first task requirement parameters, the computation task request may be assigned to an edge computing server for processing, or it may be assigned to a cloud computing center for processing. Therefore, the task requirement parameters corresponding to the first task request for the edge computing server and the task requirement parameters corresponding to the cloud computing center can be calculated. In one optional implementation provided in this application, such as... Figure 3 As shown, determining the first task requirement parameters includes steps 301 to 304:

[0060] Step 301: For each first computing task request, determine the first delay and the second delay of the first computing task request.

[0061] Wherein, the first delay represents the processing delay between the first computing task request and the edge computing server, and the second delay represents the processing delay between the first computing task request and the cloud computing center;

[0062] During implementation, for each first computing task request, when the task is computed on the edge computing server, the end-to-end latency is the sum of the wireless transmission latency and the edge computing latency, and the sum of the wireless transmission latency and the edge computing latency is taken as the first latency; when the task is processed in the cloud computing center, the end-to-end latency is mainly composed of the wireless transmission latency and the wired transmission latency, and the sum of the wireless transmission latency and the wired transmission latency is taken as the second latency.

[0063] Step 302: Determine the first index based on the comparison result between the first delay and the reference delay.

[0064] During implementation, a numerical comparison is made between the first delay and the reference delay. If the first delay is greater than the reference delay, the characterization delay is larger and the corresponding first index is smaller. If the first delay is less than or equal to the reference delay, the characterization delay is acceptable and the corresponding first index is larger.

[0065] During execution, the service quality assurance satisfaction level for tasks within the low-latency slice is defined. The latency assurance satisfaction level is 1 when the end-to-end latency of a task is not higher than a latency threshold, and 0 otherwise. This represents the latency guarantee satisfaction of the i-th task in the low-latency slice, when... This indicates that the delay in the task is guaranteed. This indicates that the task delay has exceeded the delay threshold. It is a binary function, when X is true, ,otherwise , This is the delay threshold.

[0066] The first indicator represents the satisfaction level with the service quality assurance of the first computing task and the edge computing server.

[0067] Step 303: Determine the second index based on the comparison results between the second delay and the reference delay.

[0068] During implementation, a numerical comparison between the second delay and the reference delay is performed based on the same processing procedure described above. The specific comparison process is detailed above and will not be repeated here. The second indicator represents the service quality assurance satisfaction level between the first computing task and the cloud computing center.

[0069] Step 304: Determine the first task requirement parameters based on the first indicator and the second indicator.

[0070] During implementation, the weighted average of each first computing task request with the first indicator corresponding to the edge computing server and the second indicator corresponding to the cloud computing center is calculated, and the weighted average is determined as the first task requirement parameter corresponding to the first computing task request.

[0071] In one optional implementation provided by this application, the task requirement parameters are determined by calculating the delay and comparing the delay with a reference delay, thereby improving the reliability of the determined computing task request requirements and ensuring the reliability of the allocated computing resources.

[0072] In determining the first and second delays, the respective delays can be determined based on the characteristics of the edge computing server and the cloud computing center; in one optional implementation provided in this application, such as... Figure 4 As shown, step 301 includes steps 401 to 402:

[0073] Step 401: For each first computing task request, determine the first transmission delay and the first side computing delay corresponding to the first computing task request, and add the first transmission delay and the first side computing delay together to obtain the first delay.

[0074] During implementation, for each first computing task request, the wireless transmission delay corresponding to the first computing task request and the edge computing server is determined, as well as the time required for the first computing task request to be processed by the edge computing server. The wireless transmission delay and the required time are added together to obtain the first delay.

[0075] Wherein, the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed by the edge computing server.

[0076] During execution, the wireless transmission latency is the ratio of the task size to the wireless transmission rate. Through formula express, This indicates the channel bandwidth allocated to this task. These represent the transmit power of the power IoT terminal, the channel power gain, and the Gaussian white noise interference of the channel, respectively; the edge computation latency is the ratio of the computational workload of the task to the computational resources allocated to the task by the edge server, i.e., expressed as... f and f represent the computational workload of the task and the computational resources allocated to the task by the edge server, respectively.

[0077] Step 402: Determine the second transmission delay and the third transmission delay corresponding to the first computing task request, and add the second transmission delay and the third transmission delay to obtain the second delay.

[0078] During implementation, for each first computing task request, the wireless transmission delay between the first computing task request and the cloud computing center is determined, as well as the wired transmission delay when the first computing task request is processed in the cloud computing center. The second transmission delay and the third transmission delay are added together to obtain the first delay.

[0079] The second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third transmission delay represents the wired transmission delay when the first computing task request is processed in the cloud computing center.

[0080] During execution, the wireless transmission delay is the ratio of the task size to the wireless transmission rate, and the wired transmission delay is the ratio of the task size to the bandwidth allocated to the task along the path from the base station where the task originates to the cloud computing center. A binary migration decision variable is introduced, defining that when the binary migration decision variable equals 1, the task is determined to be computed on the edge server; conversely, when the binary migration decision variable equals 0, the task is processed in the cloud computing center. Based on this, the first delay is the sum of the first transmission delay multiplied by the binary migration decision variable, and the difference between 1 and the binary migration decision variable multiplied by the first edge computation delay.

[0081] One optional implementation provided in this application determines the first delay by calculating the delay of the first computing task request on each computing device, thereby improving the reliability and accuracy of delay calculation and thus improving the accuracy of the computing resources allocated to the first computing task request.

[0082] In determining the parameters of the second task, compared to low-latency network slicing, high-throughput network slicing also requires determining bandwidth requirements; in this regard, this application provides an optional implementation method, such as... Figure 5 As shown, calculating the second task requirement parameters includes steps 501 to 504:

[0083] Step 501: For each second computing task request, determine the third latency, fourth latency, and bandwidth requirement parameters of the second computing task request.

[0084] During implementation, for each second computing task request, the latency of the corresponding edge computing server, the latency of the corresponding cloud computing center, and the bandwidth requirement parameters are calculated.

[0085] The third latency represents the processing latency between the second computing task request and the edge computing server, while the fourth latency represents the processing latency between the second computing task request and the cloud computing center.

[0086] During execution, when the second computing task request is processed on the edge server, the end-to-end latency is the sum of the wireless transmission latency and the edge computing latency, which is the third latency. When the task is processed in the cloud computing center, the end-to-end latency is mainly composed of the wireless transmission latency and the wired transmission latency, which is the fourth latency.

[0087] Step 502: Determine the third index based on the comparison results between the third delay and the reference delay.

[0088] During implementation, a numerical comparison is performed between the third delay and the reference delay to obtain the numerical comparison result. Based on the numerical comparison result, the third indicator is determined. The third indicator represents the service quality assurance satisfaction when the second computing task is computed on the edge server. The specific determination method can refer to the first indicator mentioned above, and will not be repeated here.

[0089] During execution, the end-to-end latency guarantee satisfaction is defined in high-throughput slices. The end-to-end latency guarantee satisfaction is defined as follows: when the end-to-end latency of a task is not higher than the latency threshold, the latency guarantee satisfaction is 1, otherwise it is 0.

[0090] Step 503: Determine the fourth index based on the comparison results between the fourth delay and the reference delay.

[0091] During implementation, the edge computing server determines the fourth indicator based on the comparison between the fourth latency and the reference latency. The specific comparison process can be carried out in the same way as the first and / or third indicators mentioned above, and will not be elaborated here. The fourth indicator represents the service quality assurance satisfaction when the second computing task request is processed in the cloud computing center.

[0092] Step 504: Calculate the weighted average of the third indicator, the fourth indicator, and the bandwidth requirement parameter to obtain the second task requirement parameter.

[0093] During implementation, the weights of the third indicator, the fourth indicator, and the bandwidth requirement parameter are determined, and the average of the weighted sum of the three indicators is calculated to obtain the second task requirement parameter.

[0094] During execution, the weight of bandwidth service satisfaction is no less than the weight of end-to-end latency guarantee satisfaction. The bandwidth service satisfaction is defined as follows: when the bandwidth allocated to the task is no less than the bandwidth requirement, the bandwidth service satisfaction is 1, otherwise it is 0. The service quality guarantee satisfaction of tasks in high throughput slices is defined as follows: the service quality guarantee satisfaction of tasks in high throughput slices is defined as the weighted sum of bandwidth service satisfaction and end-to-end latency guarantee satisfaction.

[0095] Among them, using This represents the service quality assurance satisfaction of the j-th task in the high-throughput slice, where... These are the weights for bandwidth service satisfaction and end-to-end latency guarantee satisfaction, respectively. Established, , These are the bandwidth service satisfaction function and the end-to-end latency guarantee satisfaction function, respectively. , It is a binary function, when X is true, ,otherwise, , .

[0096] One optional implementation provided in this application defines the service quality assurance satisfaction of tasks in a high-throughput slice as a weighted sum of bandwidth service satisfaction and end-to-end latency assurance satisfaction to represent the second requirement parameter, thereby improving the accuracy of the second requirement parameter and thus improving the reliability of the determined computing resources.

[0097] In determining the parameters of the third delay, the fourth delay, and the bandwidth requirement, this application provides an optional implementation method, such as... Figure 6 As shown, step 501 includes steps 601 to 603:

[0098] Step 601: For each second computing task request, determine the fourth transmission delay and the second side computing delay corresponding to the second computing task request, and add the fourth transmission delay and the second side computing delay to obtain the third delay.

[0099] During implementation, for each second computing task request, the edge computing server determines the wireless transmission delay corresponding to the second computing task request and the time required for processing on the edge server, and adds the wireless transmission delay and the time to obtain the third delay.

[0100] The fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server.

[0101] During execution, the wireless transmission latency is the ratio of the task size to the total wireless transmission rate, where the total wireless transmission rate is... Natural numbers can be expressed by a formula. This indicates the number of channels allocated to this task. Indicates channel bandwidth. These represent the transmit power, channel power gain, and Gaussian white noise interference of the channel, respectively. The edge computation delay is the ratio of the computational load of the task to the computational resources allocated to the task by the edge server, expressed as: , These are the task's computational workload and the computational resources allocated to the task by the edge server, respectively.

[0102] Step 602: Determine the fifth and sixth transmission delays corresponding to the second computing task request, and add the fifth and sixth transmission delays together to obtain the fourth delay.

[0103] During implementation, the edge computing server determines the wireless transmission delay corresponding to the second computing task request and the cloud computing center, as well as the wired transmission delay corresponding to the cloud computing center, and adds the wireless transmission delay and the wired transmission delay to obtain the fourth delay.

[0104] Among them, the fifth transmission delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth transmission delay represents the wired transmission delay between the second computing task request and the cloud computing center.

[0105] During execution, wireless transmission latency is the ratio of the task size to the wireless transmission rate, while wired transmission latency is the ratio of the task size to the wired transmission rate. The ratio of the bandwidth allocated to the task to the bandwidth allocated to the task along the path from the base station where the task originates to the cloud computing center; introducing binary calculations for migration decision variables. ,definition This indicates that the task is definitely computed on the edge server; otherwise, This indicates that the task is processed in the cloud computing center, and the binary migration decision variable is used in the formula. This represents the cloud-edge collaborative end-to-end latency unified model for this task; in the above formula, the symbol... This indicates that the task is a low-latency task, and the natural number is used to indicate this. Indicates the index identifier of the task.

[0106] Step 603: Determine the lower bound of the bandwidth requirement and the pre-allocated bandwidth corresponding to the second computing task request, and define the lower bound of the bandwidth requirement and the pre-allocated bandwidth as the bandwidth requirement parameters.

[0107] During implementation, a cloud-edge collaborative end-to-end bandwidth model for tasks in high-throughput network slices is determined, including the lower bound of the task's bandwidth requirement and the bandwidth allocated to the task. The lower bound of the task's bandwidth requirement is determined by the product of the task's data collection frequency and the amount of data collected per data session, i.e., using the formula... Determine the lower bound of the bandwidth requirement for the task, where the function... Indicates rounding up. This indicates the data collection frequency for this task. The bandwidth allocated to a task, representing the amount of data processed in a single run, is determined as follows: when the task is computed on an edge server, it is determined by the wireless transmission rate allocated to the task; otherwise, it is determined by the minimum value of the wireless transmission rate allocated to the task and the bandwidth allocated to that task along the path from the base station where the task originates to the cloud computing center. In other words, the bandwidth allocated to a task can be uniformly expressed as... ,in It is a minimum value function.

[0108] One optional implementation provided in this application determines the second requirement parameter by calculating the latency of tasks in a high-throughput network slice on different computing devices and the bandwidth resources required by the tasks in the high-throughput network slice. This improves the accuracy of the second requirement parameter and thus enhances the reliability of the allocated computing resources.

[0109] In determining computing resources, the resources can be determined based on the first task requirement parameters, the second task requirement parameters, and the system cost parameters to minimize system costs and thus maximize the utilization of computing resources. One optional implementation provided in this application is as follows: Figure 7 As shown, step 203 includes steps 701 to 703:

[0110] Step 701: The first task requirement parameters and the second task requirement parameters are weighted and summed to obtain the total requirement parameters.

[0111] During implementation, the edge computing server performs a weighted summation of the first task requirement parameter and the second task requirement parameter to obtain the total requirement parameter representing all computing task requests.

[0112] During the execution process, This represents the average quality of service (QoS) satisfaction level for tasks within a low-latency network slice. This represents the average quality of service (QoS) satisfaction of tasks in a high-throughput network slice. The weighted sum of these two parameters yields the total demand parameter characterizing all computing task requests.

[0113] Step 702: Obtain the system utility function by subtracting the system cost parameter from the total demand parameter.

[0114] During implementation, the system utility function is obtained by subtracting the preset system cost parameter from the total demand parameter. The system cost parameter characterizes the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of the average latency timeout penalty for each first computing task request.

[0115] During the execution process, Let the total system cost be denoted as , then the system utility is expressed as . These are the utility weights for low-latency network slices and high-throughput network slices, respectively. , These refer to the number of tasks in low-latency slices and high-throughput slices, respectively. These are the weighting coefficients for total communication resources, total computing resources, and timeout penalty, respectively. It is the total communication resources allocated by the system. It is the total computing resources allocated by the system. It is the average latency of tasks within a low-latency network slice. is the corresponding delay threshold, and max is the maximum value function.

[0116] Step 703: Solve for the maximum value of the system utility function to obtain the computing resources requested for each computing task.

[0117] During implementation, based on the regular expression of the system utility function determined above, the maximum value of the system utility function is calculated to obtain the computing resources requested for each computing task.

[0118] During execution, the system utility can be maximized as the objective, and the wireless channel resources, the bandwidth of the network transmission path from the base station to the cloud computing center, and the computing power of the edge server can be used as constraints to establish an optimization equation. The decision variables are the computational migration decision variables of the tasks in the low-latency slice and the high-throughput slice, the wireless channel resources, bandwidth resources, and computing resources allocated to the low-latency slice and the high-throughput slice. The optimization equations are as follows (see formulas (1) to (5):

[0119] P0 - Maximize: Formula (1);

[0120] Constraints: C1: Formula (2);

[0121] C2: Formula (3);

[0122] C3: Formula (4);

[0123] C4: Formula (5);

[0124] Wherein, constraint C1 is the computational migration decision variable for tasks in low-latency slices and high-throughput slices; C2 is the wireless channel resource constraint. C1 represents the number of available wireless channels; C2 represents the bandwidth limitation on the path from the base station to the cloud computing center. C4 represents the available bandwidth of the path; C5 represents the limitation on the computing resources of the edge server. It is the available computing resources of the edge server.

[0125] Furthermore, the optimization equation can be solved using problem decomposition and iterative algorithms to obtain the optimal solution, which includes the best computational migration decisions for tasks in low-latency and high-throughput slices. Optimal wireless channel resources allocated to the task ,bandwidth and computing resource strategy Compute migration decisions characterize the decision of whether to allocate computing task requests to edge computing servers or to cloud computing centers.

[0126] One optional implementation provided in this application ensures that the computing resources allocated to each computing task are minimized by establishing a system utility expression and solving for the maximum value, thereby improving resource utilization.

[0127] In determining computing resources, the system utility function can be solved through various network collaboration methods to obtain accurate computing resources; in one optional implementation method provided in this application, such as Figure 8 As shown, step 703 includes step 801:

[0128] Step 801: Iteratively solve the neural network and gradient network according to the system utility function, and obtain the computing resources requested for each computing task when the neural network and gradient network converge.

[0129] During implementation, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request. The computing device information and the system utility function are then input into the gradient network to obtain the computing resources, and an iterative solution is performed. That is, the computing resources and the system utility function are continuously input into the neural network. Specifically, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network. The system utility function and the computing device information are then input into the gradient network to obtain the computing resources corresponding to each computing task request.

[0130] It should be noted that neural networks can adjust their parameters based on the output of the previous round, and similarly, gradient networks can also adjust their parameters based on the output of the previous round, so as to continuously iterate the network during application.

[0131] During execution, the problem is decomposed into a computation migration decision optimization subproblem and a computation-network collaborative resource allocation subproblem. The computation migration decision optimization subproblem, given slice resources, is an optimization equation with the computation migration decision variables of tasks in low-latency and high-throughput slices as decision variables, aiming to maximize slice utility. The computation-network collaborative resource allocation subproblem, given the computation migration decisions of two network slices, is an optimization equation with the allocation of wireless channel resources, bandwidth resources, and computational resources as decision variables, aiming to maximize system utility. In other words, the computation migration decision optimization subproblem is: P1 - Maximize: The constraints are C1-1 to C1-5; P2-Maximize: The constraints are C2-1 to C2-5;

[0132] C1-1;

[0133] C1-2;

[0134] C1-3;

[0135] C1-4;

[0136] C1-5;

[0137] C2-1;

[0138] C2-2;

[0139] C2-3;

[0140] C2-4;

[0141] C2-5;

[0142] Where P1 is the computational migration decision optimization subproblem for low-latency slices, P2 is the computational migration decision optimization subproblem for high-throughput slices, and the right sides of the equations C1-3 and C2-3 come from the solution of the P3 subproblem.

[0143] The subproblem of collaborative resource allocation in computing networks is:

[0144] P3 - Maximize: Restrictions: , And C2-C4; where the right side of the equation in C3-1 is the solution from subproblems P1 and P2;

[0145] Then, the solutions to the two sub-problems are solved using an iterative algorithm. In each iteration, the computational transfer decision optimization sub-problems of low-latency slices and high-throughput slices are solved in parallel using neural networks, i.e., solving P1 and P2. The computational network collaborative resource allocation sub-problem is solved using a gradient network, i.e., solving P3. The computational transfer decisions of low-latency slices and high-throughput slices output by the neural networks in parallel are aggregated and used as the input for the next round of learning of the gradient network. After the gradient network outputs the resource allocation strategy, the slice resources of low-latency slices and high-throughput slices are determined according to the resource allocation results, and then used as the input for the next round of learning of the two neural networks respectively.

[0146] When the strategies output by neural networks and gradient networks fail to increase system utility When the iteration stops, the outputs of the current neural network and gradient network are aggregated and used as the best solution to the problem. The solution includes the best computational migration decision for the task in the low-latency slice and the high-throughput slice, the best wireless channel resources allocated to the task, and the bandwidth and computational resource strategy.

[0147] The specific method for determining the slice resources of low-latency slices and high-throughput slices based on the resource allocation results is as follows: Available wireless channel resources are allocated to low-latency slices and high-throughput slices according to the ratio of the sum of wireless channel resources allocated to tasks in low-latency slices to the sum of wireless channel resources allocated to tasks in high-throughput slices in the current resource allocation strategy; the bandwidth of the path from the base station to the cloud computing center is allocated to low-latency slices and high-throughput slices according to the ratio of the sum of bandwidth allocated to tasks in low-latency slices to the sum of bandwidth allocated to tasks in high-throughput slices in the current resource allocation strategy; and the computing resources of the side servers are allocated to low-latency slices and high-throughput slices according to the ratio of the sum of edge server computing resources allocated to tasks in low-latency slices to the sum of edge server computing resources allocated to tasks in high-throughput slices in the current resource allocation strategy.

[0148] One optional implementation provided in this application improves the reliability and accuracy of the computing resources obtained by decomposing the maximization computation problem into two sub-problems for separate computation, thereby improving resource utilization.

[0149] If the neural network does not converge, it still needs to adjust its parameters in each round; in one optional implementation provided in this application, the neural network includes a deep Q-network, such as... Figure 9 As shown, the parameter tuning process of the neural network includes steps 901 to 902:

[0150] Step 901: For each first computing task request in each computing task request, the first task requirement parameter corresponding to the first computing task request is used as the first state space, the computing device information corresponding to the first computing task request is used as the first action space, and the system utility function corresponding to the first computing task request is used as the first reward function.

[0151] During implementation, for each first computing task request in each computing task request, the deep Q network uses the first task requirement parameters corresponding to the first computing task request as the first state space, the computing device information corresponding to the first computing task request as the first action space, and the system utility function corresponding to the first computing task request as the first reward function.

[0152] During execution, the current task information and slice resource status are taken as the current state; that is, let the state at training time t be represented as:

[0153] The state space is the set of all states; the combination of computational transfer decisions for in-chip tasks is taken as actions, i.e. The variant of the system utility function, i.e., the system utility function of the task set containing only the intra-slice tasks, is used as the reward function for this slice; that is, the reward function at training time t is... The state-action value function is defined as the Q-value of choosing an action in a given state in an online Q-network, i.e., ,in, The parameters of the online Q-network are represented; the loss function is defined as follows: ,in, It is an instant reward value. It is a discount factor. These are the state and movement space at the next training moment, respectively. It is the Q-value function of the target Q-network. These are the parameters of the target Q-network; the parameters of the target Q-network are set to be updated at regular training steps, that is, when t is an integer multiple of T, let , where T is the training step size of the interval.

[0154] Step 902: Determine a first loss value based on the first state space, the first action space, and the first reward function, and adjust the parameters of the neural network based on the first loss value.

[0155] During implementation, a first loss value is determined based on a first state space, a first action space, and a first reward function. The parameters of the neural network are then adjusted based on the first loss value to obtain the neural network after this round of parameter tuning.

[0156] Furthermore, the method for solving computational transfer decisions in low-latency slices and high-throughput slices based on the trained deep Q-network is as follows: Let If the current state of the low-latency slice is indicated, the computation migration strategy for that slice is selected according to the following rules: .

[0157] One optional implementation provided in this application determines the neural network as a deep Q-network. By leveraging the characteristics of the deep Q-network for iterative solution and transfer strategy calculation, the accuracy of the transfer strategy is improved, thereby enhancing the reliability of the computing resources obtained from the calculation.

[0158] If the gradient network has not converged, the gradient network still needs to be adjusted in each round; in one optional implementation provided in this application, such as Figure 10 As shown, the parameter tuning process of the gradient network includes steps 1001 to 1002:

[0159] Step 1001: For each second computing task request in each computing task request, take the second task requirement parameters corresponding to the second computing task request as the second state space, take the computing power resources corresponding to the second computing task request as the second action space, and take the system utility function corresponding to the second computing task request as the second reward function.

[0160] During implementation, for each second computing task request in each computing task request, the second task requirement parameters corresponding to the second computing task request are used as the second state space, the computing resources corresponding to the second computing task request are used as the second action space, and the system utility function corresponding to the second computing task request is used as the second reward function.

[0161] During execution, the current task information, computational migration decisions, and available system resources are used as the current state; that is, let the state at training time t be represented as... ={ , The state space is the set of all states; the action space is the combination of allocation strategies for wireless channel resources, bandwidth resources, and computing resources, that is, the action at time t can be represented as... ={ , }; The system utility is used as the reward function, that is, The deep deterministic policy gradient network comprises an online executor network, an online evaluation network, a target executor network, and a target evaluation network, with network parameters respectively using... In other words, the policy function is defined as a function in an online actor network that maps the current state to behavior; that is, using... Denotes the policy function, where, It is the current state. These are the parameters of the online actor network. The state-action value function is defined as the Q-value of choosing an action in a given state within the target actor network, i.e., The mathematical expectation of long-term discount cumulative rewards is defined as: ,in, It is a discount factor. It is the instantaneous reward function at the corresponding time point; the loss function of the online evaluation network is defined as: ,in, It refers to the state and action at time t+1.

[0162] Step 1002: Determine the second loss value based on the second state space, the second action space, and the second reward function, and adjust the parameters of the gradient network based on the second loss value.

[0163] During implementation, a second loss value is determined based on the second state space, the second action space, and the second reward function. The parameters of the gradient network are then adjusted based on the second loss value to obtain the gradient network for the current round.

[0164] Furthermore, the method for solving the cloud-edge collaborative resource allocation strategy based on the trained deep deterministic policy gradient network is as follows: Let Indicates the current state. The computational migration strategy after the solution aggregation of subproblems P1 and P2 in the current state is given, and the action is selected according to the following rules: Assumption:

[0165] The reward obtained based on the above actions is Therefore, the optimal solution for P3 is obtained:

[0166] Combining the optimized solutions of P1 and P2, we obtain the optimized solution for problem P0:

[0167] { and optimal system utility .

[0168] One optional implementation method provided in this application solves the computational resources through a gradient network, which improves the reliability of the computational resources and thus improves the resource utilization of the system.

[0169] In one embodiment, see Figure 11 The diagram illustrates a flowchart of a resource allocation method provided in an embodiment of this application, which can be applied to... Figure 1 In the edge computing server shown. For example... Figure 11 As shown, the resource allocation method may include the following steps:

[0170] Step 1101: Receive at least one computing task request sent by a power IoT device, and allocate each computing task request to a low-latency network slice and a high-throughput network slice.

[0171] Step 1102: For each first computing task request, determine the first delay and the second delay of the first computing task request.

[0172] Step 1103: Based on the comparison result between the first delay and the reference delay, determine the first index, and based on the comparison result between the second delay and the reference delay, determine the second index.

[0173] Step 1104: For each second computing task request, determine the third latency, fourth latency, and bandwidth requirement parameters of the second computing task request.

[0174] Step 1105: Based on the comparison results between the third delay and the reference delay, determine the third index, and based on the comparison results between the fourth delay and the reference delay, determine the fourth index.

[0175] Step 1106: Calculate the weighted average of the third indicator, the fourth indicator, and the bandwidth requirement parameter to obtain the second task requirement parameter.

[0176] Step 1107: The first task requirement parameter and the second task requirement parameter are weighted and summed to obtain the total requirement parameter, and the system utility function is obtained by subtracting the system cost parameter from the total requirement parameter.

[0177] Step 1108: Iteratively solve the neural network and gradient network according to the system utility function, and obtain the computing resources allocated to each computing task when the neural network and gradient network converge.

[0178] It should be noted that any one or more of steps 1101 to 1108 can be combined to form a new implementation method according to the needs of implementation and deployment. Furthermore, any one or more technical features in the technical solution composed of steps 1101 to 1108 can also be combined to form a new implementation method according to the actual deployment needs, or technical features in one or more optional implementations provided by one or more of the above embodiments can be combined to form a new implementation method. These will not be elaborated on here.

[0179] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0180] Based on the same inventive concept, this application also provides a resource allocation apparatus for implementing the resource allocation method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more resource allocation apparatus embodiments provided below can be found in the limitations of the resource allocation method described above, and will not be repeated here.

[0181] In one exemplary embodiment, such as Figure 12 As shown, a resource allocation device is provided, including: a task request allocation module 1201, a demand parameter calculation module 1202, and a computing power resource determination module 1203, wherein: the task request allocation module 1201 is used to receive at least one computing task request sent by a power Internet of Things (IoT) device, and allocate each computing task request to a low-latency network slice and a high-throughput network slice; the demand parameter calculation module 1202 is used to calculate a first task demand parameter of a first computing task request included in a low-latency network slice, and calculate a second task demand parameter of a second computing task request included in a high-throughput network slice; the computing power resource determination module 1203 is used to determine the computing power resources allocated to each computing task request based on the first task demand parameter and the second task demand parameter.

[0182] In one embodiment, the computing resource determination module 1203 includes a demand determination unit, a utility function determination unit, and a computing resource solution unit, wherein: the demand determination unit is used to perform a weighted summation of the first task demand parameters and the second task demand parameters to obtain a total demand parameter; the utility function determination unit is used to obtain a system utility function by subtracting a system cost parameter from the total demand parameter, wherein the system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each of the computing task requests; and the computing resource solution unit is used to solve for the maximum value of the system utility function to obtain the computing resources allocated to each of the computing task requests.

[0183] In one embodiment, the computing power resource solving unit includes an iterative solving unit, which is used to iteratively solve the neural network and the gradient network according to the system utility function, and obtain the computing power resources requested to be allocated for each of the computing tasks when the neural network and the gradient network converge.

[0184] In one embodiment, the apparatus further includes a first determining unit and a first adjusting unit, wherein: the first determining unit is configured to, for each of the computing task requests, take the first task requirement parameter corresponding to the first computing task request as a first state space, take the computing device information corresponding to the first computing task request as a first action space, and take the system utility function corresponding to the first computing task request as a first reward function; the first adjusting unit is configured to determine a first loss value based on the first state space, the first action space and the first reward function, and adjust the parameters of the neural network based on the first loss value.

[0185] In one embodiment, the apparatus further includes a second determining unit and a second adjusting unit, wherein: the second determining unit is configured to, for each of the second computing task requests in each of the computing task requests, use the second task requirement parameters corresponding to the second computing task request as a second state space, use the computing power resources corresponding to the second computing task request as a second action space, and use the system utility function corresponding to the second computing task request as a second reward function; the second adjusting unit is configured to determine a second loss value based on the second state space, the second action space, and the second reward function, and adjust the parameters of the gradient network based on the second loss value.

[0186] In one embodiment, the requirement parameter calculation module 1202 includes a first latency determination unit, a first indicator determination unit, a second indicator determination unit, and a first task requirement determination unit, wherein: the first latency determination unit is used to determine a first latency and a second latency for each first computing task request, wherein the first latency represents the processing latency of the first computing task request corresponding to the edge computing server, and the second latency represents the processing latency of the first computing task request corresponding to the cloud computing center; the first indicator determination unit is used to determine a first indicator based on the comparison result between the first latency and a reference latency, wherein the first indicator represents the service quality assurance satisfaction of the first computing task corresponding to the edge computing server; the second indicator determination unit is used to determine a second indicator based on the comparison result between the second latency and the reference latency, wherein the second indicator represents the service quality assurance satisfaction of the first computing task corresponding to the cloud computing center; and the first task requirement determination unit is used to determine the first indicator and the second indicator as the first task requirement parameters.

[0187] In one embodiment, the first delay determination unit includes a first delay determination subunit and a second delay determination subunit, wherein: the first delay determination subunit is used to determine, for each first computing task request, a first transmission delay and a first edge computing delay corresponding to the first computing task request, and add the first transmission delay and the first edge computing delay to obtain the first delay, wherein the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed by the edge computing server; the second delay determination subunit is used to determine a second transmission delay and a third transmission delay corresponding to the first computing task request, and add the second delay and the third delay to obtain the second delay, wherein the second delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third delay represents the wired transmission delay when the first computing task request is processed by the cloud computing center.

[0188] In one embodiment, the requirement parameter calculation module 1202 includes a second latency determination unit, a third indicator determination unit, a fourth indicator determination unit, and a second task requirement determination unit, wherein: the second latency determination unit is used to determine a third latency, a fourth latency, and bandwidth requirement parameters for each second computing task request, wherein the third latency represents the processing latency of the second computing task request corresponding to the edge computing server, and the fourth latency represents the processing latency of the second computing task request corresponding to the cloud computing center; the third indicator determination unit is used to determine a third indicator based on the comparison result between the third latency and the reference latency, wherein the third indicator represents the service quality assurance satisfaction of the second computing task when computing on the edge server; the fourth indicator determination unit is used to determine a fourth indicator based on the comparison result between the fourth latency and the reference latency, wherein the fourth indicator represents the service quality assurance satisfaction of the second computing task request when computing on the cloud computing center; and the second task requirement determination unit is used to calculate the weighted average sum of the third indicator, the fourth indicator, and the bandwidth requirement parameters to obtain the second task requirement parameters.

[0189] In one embodiment, the second latency determination unit includes a third latency determination subunit, a fourth latency determination subunit, and a bandwidth requirement determination unit, wherein: the third latency determination subunit is used to determine, for each second computing task request, a fourth transmission latency and a second edge computing latency corresponding to the second computing task request, and add the fourth transmission latency and the second edge computing latency to obtain the third latency, wherein the fourth transmission latency represents the wireless transmission latency between the second computing task request and the edge computing server, and the second edge computing latency represents the time required for the second computing task request to be processed by the edge server; the fourth latency determination subunit is used to determine a fifth transmission latency and a sixth transmission latency corresponding to the second computing task request, and add the fifth latency and the sixth latency to obtain the fourth latency, wherein the fifth latency represents the wireless transmission latency between the second computing task request and the cloud computing center, and the sixth latency represents the wired transmission latency between the second computing task request and the cloud computing center; the bandwidth requirement determination unit is used to determine the lower bound of the bandwidth requirement and the pre-allocated bandwidth corresponding to the second computing task request, and determine the lower bound of the bandwidth requirement and the pre-allocated bandwidth as the bandwidth requirement parameter.

[0190] Each module in the aforementioned resource allocation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0191] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 13 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores resource allocation data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a resource allocation method.

[0192] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0193] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: receiving at least one computing task request sent by a power Internet of Things (IoT) device; allocating each computing task request to a low-latency network slice and a high-throughput network slice; calculating a first task requirement parameter for a first computing task request included in a low-latency network slice, and calculating a second task requirement parameter for a second computing task request included in a high-throughput network slice; and determining the computing resources allocated to each computing task request based on the first task requirement parameter and the second task requirement parameter.

[0194] In one embodiment, when the processor executes the computer program, it further performs the following steps: weighted summing of the first task requirement parameter and the second task requirement parameter to obtain the total requirement parameter; subtracting the system cost parameter from the total requirement parameter to obtain the system utility function, where the system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each computing task request; and solving for the maximum value of the system utility function to obtain the computing resources allocated to each computing task request.

[0195] In one embodiment, when the processor executes the computer program, it further performs the following steps: iteratively solving the neural network and gradient network according to the system utility function, and obtaining the computing resources allocated to each computing task request when the neural network and gradient network converge; wherein, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network, and the system utility function and the computing device information are input into the gradient network to obtain the computing resources corresponding to each computing task request.

[0196] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request in each computing task request, it uses the first task requirement parameter corresponding to the first computing task request as a first state space, the computing device information corresponding to the first computing task request as a first action space, and the system utility function corresponding to the first computing task request as a first reward function; it determines a first loss value based on the first state space, the first action space, and the first reward function, and adjusts the parameters of the neural network based on the first loss value.

[0197] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request in each computing task request, the second task requirement parameter corresponding to the second computing task request is used as a second state space, the computing power resources corresponding to the second computing task request are used as a second action space, and the system utility function corresponding to the second computing task request is used as a second reward function; a second loss value is determined based on the second state space, the second action space, and the second reward function, and the parameters of the gradient network are adjusted based on the second loss value.

[0198] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first latency and a second latency of the first computing task request, wherein the first latency characterizes the processing latency of the first computing task request corresponding to the edge computing server, and the second latency characterizes the processing latency of the first computing task request corresponding to the cloud computing center; determining a first index based on a comparison between the first latency and a reference latency, wherein the first index characterizes the service quality assurance satisfaction of the first computing task corresponding to the edge computing server; determining a second index based on a comparison between the second latency and the reference latency, wherein the second index characterizes the service quality assurance satisfaction of the first computing task corresponding to the cloud computing center; and determining first task requirement parameters according to the first index and the second index.

[0199] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first transmission delay and a first edge computing delay corresponding to the first computing task request, and adding the first transmission delay and the first edge computing delay to obtain a first delay, wherein the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed on the edge computing server; determining a second transmission delay and a third transmission delay corresponding to the first computing task request, and adding the second delay and the third delay to obtain a second delay, wherein the second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third delay represents the wired transmission delay when the first computing task request is processed on the cloud computing center.

[0200] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a third latency, a fourth latency, and bandwidth requirement parameters for the second computing task request, wherein the third latency characterizes the processing latency corresponding to the second computing task request and the edge computing server, and the fourth latency characterizes the processing latency corresponding to the second computing task request and the cloud computing center; based on the comparison result between the third latency and the reference latency, determining a third index, wherein the third index characterizes the service quality assurance satisfaction when the second computing task is computed on the edge server; based on the comparison result between the fourth latency and the reference latency, determining a fourth index, wherein the fourth index characterizes the service quality assurance satisfaction when the second computing task request is computed in the cloud computing center; and calculating the weighted average sum of the third index, the fourth index, and the bandwidth requirement parameters to obtain the second task requirement parameters.

[0201] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a fourth transmission delay and a second edge computing delay corresponding to the second computing task request, and adding the fourth transmission delay and the second edge computing delay to obtain a third delay, wherein the fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server; determining a fifth transmission delay and a sixth transmission delay corresponding to the second computing task request, and adding the fifth delay and the sixth delay to obtain a fourth delay, wherein the fifth delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth delay represents the wired transmission delay between the second computing task request and the cloud computing center; determining a lower bound of bandwidth requirements and a pre-allocated bandwidth corresponding to the second computing task request, and determining the lower bound of bandwidth requirements and the pre-allocated bandwidth as bandwidth requirement parameters.

[0202] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon. When executed by a processor, the computer program performs the following steps: receiving at least one computing task request sent by a power Internet of Things (IoT) device; allocating each computing task request to a low-latency network slice and a high-throughput network slice; calculating a first task requirement parameter for a first computing task request included in the low-latency network slice, and calculating a second task requirement parameter for a second computing task request included in the high-throughput network slice; and determining the computing resources allocated to each computing task request based on the first and second task requirement parameters.

[0203] In one embodiment, when the processor executes the computer program, it further performs the following steps: weighted summing of the first task requirement parameter and the second task requirement parameter to obtain the total requirement parameter; subtracting the system cost parameter from the total requirement parameter to obtain the system utility function, where the system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each computing task request; and solving for the maximum value of the system utility function to obtain the computing resources allocated to each computing task request.

[0204] In one embodiment, when the processor executes the computer program, it further performs the following steps: iteratively solving the neural network and gradient network according to the system utility function, and obtaining the computing resources allocated to each computing task request when the neural network and gradient network converge; wherein, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network, and the system utility function and the computing device information are input into the gradient network to obtain the computing resources corresponding to each computing task request.

[0205] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request in each computing task request, it uses the first task requirement parameter corresponding to the first computing task request as a first state space, the computing device information corresponding to the first computing task request as a first action space, and the system utility function corresponding to the first computing task request as a first reward function; it determines a first loss value based on the first state space, the first action space, and the first reward function, and adjusts the parameters of the neural network based on the first loss value.

[0206] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request in each computing task request, the second task requirement parameter corresponding to the second computing task request is used as a second state space, the computing power resources corresponding to the second computing task request are used as a second action space, and the system utility function corresponding to the second computing task request is used as a second reward function; a second loss value is determined based on the second state space, the second action space, and the second reward function, and the parameters of the gradient network are adjusted based on the second loss value.

[0207] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first latency and a second latency of the first computing task request, wherein the first latency characterizes the processing latency of the first computing task request corresponding to the edge computing server, and the second latency characterizes the processing latency of the first computing task request corresponding to the cloud computing center; determining a first index based on a comparison between the first latency and a reference latency, wherein the first index characterizes the service quality assurance satisfaction of the first computing task corresponding to the edge computing server; determining a second index based on a comparison between the second latency and the reference latency, wherein the second index characterizes the service quality assurance satisfaction of the first computing task corresponding to the cloud computing center; and determining first task requirement parameters according to the first index and the second index.

[0208] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first transmission delay and a first edge computing delay corresponding to the first computing task request, and adding the first transmission delay and the first edge computing delay to obtain a first delay, wherein the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed on the edge computing server; determining a second transmission delay and a third transmission delay corresponding to the first computing task request, and adding the second delay and the third delay to obtain a second delay, wherein the second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third delay represents the wired transmission delay when the first computing task request is processed on the cloud computing center.

[0209] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a third latency, a fourth latency, and bandwidth requirement parameters for the second computing task request, wherein the third latency characterizes the processing latency corresponding to the second computing task request and the edge computing server, and the fourth latency characterizes the processing latency corresponding to the second computing task request and the cloud computing center; based on the comparison result between the third latency and the reference latency, determining a third index, wherein the third index characterizes the service quality assurance satisfaction when the second computing task is computed on the edge server; based on the comparison result between the fourth latency and the reference latency, determining a fourth index, wherein the fourth index characterizes the service quality assurance satisfaction when the second computing task request is computed in the cloud computing center; and calculating the weighted average sum of the third index, the fourth index, and the bandwidth requirement parameters to obtain the second task requirement parameters.

[0210] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a fourth transmission delay and a second edge computing delay corresponding to the second computing task request, and adding the fourth transmission delay and the second edge computing delay to obtain a third delay, wherein the fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server; determining a fifth transmission delay and a sixth transmission delay corresponding to the second computing task request, and adding the fifth delay and the sixth delay to obtain a fourth delay, wherein the fifth delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth delay represents the wired transmission delay between the second computing task request and the cloud computing center; determining a lower bound of bandwidth requirements and a pre-allocated bandwidth corresponding to the second computing task request, and determining the lower bound of bandwidth requirements and the pre-allocated bandwidth as bandwidth requirement parameters.

[0211] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: receiving at least one computing task request sent by a power Internet of Things (IoT) device; allocating each computing task request to a low-latency network slice and a high-throughput network slice; calculating a first task requirement parameter for a first computing task request included in the low-latency network slice, and calculating a second task requirement parameter for a second computing task request included in the high-throughput network slice; and determining the computing resources allocated to each computing task request based on the first and second task requirement parameters.

[0212] In one embodiment, when the processor executes the computer program, it further performs the following steps: weighted summing of the first task requirement parameter and the second task requirement parameter to obtain the total requirement parameter; subtracting the system cost parameter from the total requirement parameter to obtain the system utility function, where the system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each computing task request; and solving for the maximum value of the system utility function to obtain the computing resources allocated to each computing task request.

[0213] In one embodiment, when the processor executes the computer program, it further performs the following steps: iteratively solving the neural network and gradient network according to the system utility function, and obtaining the computing resources allocated to each computing task request when the neural network and gradient network converge; wherein, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network, and the system utility function and the computing device information are input into the gradient network to obtain the computing resources corresponding to each computing task request.

[0214] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request in each computing task request, it uses the first task requirement parameter corresponding to the first computing task request as a first state space, the computing device information corresponding to the first computing task request as a first action space, and the system utility function corresponding to the first computing task request as a first reward function; it determines a first loss value based on the first state space, the first action space, and the first reward function, and adjusts the parameters of the neural network based on the first loss value.

[0215] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request in each computing task request, the second task requirement parameter corresponding to the second computing task request is used as a second state space, the computing power resources corresponding to the second computing task request are used as a second action space, and the system utility function corresponding to the second computing task request is used as a second reward function; a second loss value is determined based on the second state space, the second action space, and the second reward function, and the parameters of the gradient network are adjusted based on the second loss value.

[0216] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first latency and a second latency of the first computing task request, wherein the first latency characterizes the processing latency of the first computing task request corresponding to the edge computing server, and the second latency characterizes the processing latency of the first computing task request corresponding to the cloud computing center; determining a first index based on a comparison between the first latency and a reference latency, wherein the first index characterizes the service quality assurance satisfaction of the first computing task corresponding to the edge computing server; determining a second index based on a comparison between the second latency and the reference latency, wherein the second index characterizes the service quality assurance satisfaction of the first computing task corresponding to the cloud computing center; and determining first task requirement parameters according to the first index and the second index.

[0217] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each first computing task request, determining a first transmission delay and a first edge computing delay corresponding to the first computing task request, and adding the first transmission delay and the first edge computing delay to obtain a first delay, wherein the first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed on the edge computing server; determining a second transmission delay and a third transmission delay corresponding to the first computing task request, and adding the second delay and the third delay to obtain a second delay, wherein the second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third delay represents the wired transmission delay when the first computing task request is processed on the cloud computing center.

[0218] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a third latency, a fourth latency, and bandwidth requirement parameters for the second computing task request, wherein the third latency characterizes the processing latency corresponding to the second computing task request and the edge computing server, and the fourth latency characterizes the processing latency corresponding to the second computing task request and the cloud computing center; based on the comparison result between the third latency and the reference latency, determining a third index, wherein the third index characterizes the service quality assurance satisfaction when the second computing task is computed on the edge server; based on the comparison result between the fourth latency and the reference latency, determining a fourth index, wherein the fourth index characterizes the service quality assurance satisfaction when the second computing task request is computed in the cloud computing center; and calculating the weighted average sum of the third index, the fourth index, and the bandwidth requirement parameters to obtain the second task requirement parameters.

[0219] In one embodiment, when the processor executes the computer program, it further performs the following steps: for each second computing task request, determining a fourth transmission delay and a second edge computing delay corresponding to the second computing task request, and adding the fourth transmission delay and the second edge computing delay to obtain a third delay, wherein the fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server; determining a fifth transmission delay and a sixth transmission delay corresponding to the second computing task request, and adding the fifth delay and the sixth delay to obtain a fourth delay, wherein the fifth delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth delay represents the wired transmission delay between the second computing task request and the cloud computing center; determining a lower bound of bandwidth requirements and a pre-allocated bandwidth corresponding to the second computing task request, and determining the lower bound of bandwidth requirements and the pre-allocated bandwidth as bandwidth requirement parameters.

[0220] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0221] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0222] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0223] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A resource allocation method, characterized in that, Applied to edge computing servers, the method includes: Receive at least one computing task request sent by a power Internet of Things (IoT) device, and allocate each computing task request to a low-latency network slice and a high-throughput network slice; Calculate the first task requirement parameters for the first computing task request included in the low-latency network slice, and calculate the second task requirement parameters for the second computing task request included in the high-throughput network slice; The computing resources requested to be allocated to each computing task are determined based on the first task requirement parameters and the second task requirement parameters.

2. The method according to claim 1, characterized in that, The computing resources allocated to each computing task request based on the first task requirement parameters and the second task requirement parameters include: The first task requirement parameters and the second task requirement parameters are weighted and summed to obtain the total requirement parameters. The system utility function is obtained by subtracting the system cost parameter from the total demand parameter. The system cost parameter is used to characterize the weighted sum of communication resources, the weighted sum of computing resources, and the weighted sum of average latency timeout penalties corresponding to each of the first computing task requests. The maximum value of the system utility function is calculated to obtain the computing resources requested for each computing task.

3. The method according to claim 2, characterized in that, The step of maximizing the system utility function to obtain the computing resources requested for each computing task includes: The neural network and gradient network are iteratively solved according to the system utility function, and the computing resources requested for each computing task are obtained when the neural network and gradient network converge. Specifically, for one solution process in the iterative solution, the system utility function is input into the neural network to obtain the computing device information corresponding to each computing task request output by the neural network, and the system utility function and the computing device information are input into the gradient network to obtain the computing power resources corresponding to each computing task request.

4. The method according to claim 3, characterized in that, In the event that the neural network fails to converge, the method further includes: For each of the first computing task requests in each of the computing task requests, the first task requirement parameters corresponding to the first computing task request are used as the first state space, the computing device information corresponding to the first computing task request is used as the first action space, and the system utility function corresponding to the first computing task request is used as the first reward function. A first loss value is determined based on the first state space, the first action space, and the first reward function, and the parameters of the neural network are adjusted based on the first loss value.

5. The method according to claim 3, characterized in that, In the event that the gradient network fails to converge, the method further includes: For each of the second computing task requests in the aforementioned computing task requests, the second task requirement parameters corresponding to the second computing task request are used as the second state space, the computing power resources corresponding to the second computing task request are used as the second action space, and the system utility function corresponding to the second computing task request is used as the second reward function. A second loss value is determined based on the second state space, the second action space, and the second reward function, and the parameters of the gradient network are adjusted based on the second loss value.

6. The method according to claim 1, characterized in that, The calculation of the first task requirement parameters for the first computing task request included in the low-latency network slice includes: For each of the first computing task requests, a first latency and a second latency are determined, wherein the first latency represents the processing latency of the first computing task request corresponding to the edge computing server, and the second latency represents the processing latency of the first computing task request corresponding to the cloud computing center; Based on the comparison result between the first delay and the reference delay, a first indicator is determined, which represents the service quality assurance satisfaction of the first computing task and the edge computing server. Based on the comparison between the second delay and the reference delay, a second indicator is determined, which represents the service quality assurance satisfaction of the first computing task and the cloud computing center. The first task requirement parameters are determined based on the first indicator and the second indicator.

7. The method according to claim 6, characterized in that, For each of the first computing task requests, determining the first latency and the second latency of the first computing task request includes: For each first computing task request, a first transmission delay and a first edge computing delay corresponding to the first computing task request are determined, and the first transmission delay and the first edge computing delay are added together to obtain the first delay. The first transmission delay represents the wireless transmission delay between the first computing task request and the edge computing server, and the first edge computing delay represents the time required for the first computing task request to be processed by the edge computing server. The second transmission delay and the third transmission delay corresponding to the first computing task request are determined, and the second transmission delay and the third transmission delay are added together to obtain the second delay. The second transmission delay represents the wireless transmission delay between the first computing task request and the cloud computing center, and the third transmission delay represents the wired transmission delay when the first computing task request is processed in the cloud computing center.

8. The method according to claim 1, characterized in that, The second task requirement parameters for calculating the second computing task request included in the high-throughput network slice include: For each second computing task request, a third latency, a fourth latency, and a bandwidth requirement parameter are determined for the second computing task request. The third latency represents the processing latency of the second computing task request corresponding to the edge computing server, and the fourth latency represents the processing latency of the second computing task request corresponding to the cloud computing center. Based on the comparison results between the third delay and the reference delay, a third indicator is determined, which characterizes the service quality assurance satisfaction when the second computing task is computed on the edge server. Based on the comparison between the fourth delay and the reference delay, a fourth indicator is determined, which characterizes the service quality assurance satisfaction when the second computing task request is computed in the cloud computing center. The second task requirement parameter is obtained by calculating the weighted average of the third indicator, the fourth indicator, and the bandwidth requirement parameter.

9. The method according to claim 8, characterized in that, For each of the second computing task requests, determining the third latency, fourth latency, and bandwidth requirement parameters of the second computing task request includes: For each second computing task request, a fourth transmission delay and a second edge computing delay corresponding to the second computing task request are determined, and the fourth transmission delay and the second edge computing delay are added together to obtain the third delay. The fourth transmission delay represents the wireless transmission delay between the second computing task request and the edge computing server, and the second edge computing delay represents the time required for the second computing task request to be processed by the edge server. The fifth transmission delay and the sixth transmission delay corresponding to the second computing task request are determined, and the fifth transmission delay and the sixth transmission delay are added together to obtain the fourth delay. The fifth transmission delay represents the wireless transmission delay between the second computing task request and the cloud computing center, and the sixth transmission delay represents the wired transmission delay between the second computing task request and the cloud computing center. Determine the lower bound of the bandwidth requirement and the pre-allocated bandwidth corresponding to the second computing task request, and define the lower bound of the bandwidth requirement and the pre-allocated bandwidth as the bandwidth requirement parameter.

10. A resource allocation device, characterized in that, The device includes: The task request allocation module is used to receive at least one computing task request sent by a power Internet of Things device and allocate each computing task request to a low-latency network slice and a high-throughput network slice. The requirement parameter calculation module is used to calculate the first task requirement parameters of the first computing task request included in the low-latency network slice, and to calculate the second task requirement parameters of the second computing task request included in the high-throughput network slice. The computing resource determination module is used to determine the computing resources requested to be allocated to each of the computing tasks based on the first task requirement parameters and the second task requirement parameters.

Images