Computing power resource access method of intelligent computing center

Abstract

The invention discloses a computing power resource access method of an intelligent computing center, which relates to the technical field of computing power resource access, and comprises the following steps: identifying the identity of a user, and calculating the computing power demand of the current computing power resource access task of the user; real-time redundant computing power is calculated, the unit maximum computing power access amount corresponding to the computing power resource access permission level of the current computing power resource access task of the user is obtained, and auxiliary access analysis early warning is sent out; after the auxiliary access analysis early warning is received, calculating a real-time computing power abnormal coefficient, and sending a computing power correction instruction to the outside; after a computing power correction instruction is received, real-time redundant computing power and a real-time computing power abnormal coefficient are obtained, the corrected redundant computing power is calculated, the computing power requirement of the current computing power resource access task of the user is combined, the current computing power resource access permissible computing power and the permissible computing power access duration of the user are output for access execution, and the access execution is performed in the computing power requirement peak period. And the system dynamically distributes and corrects redundant computing power to ensure that training tasks are completed on time.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power resource access, and specifically provides a method for accessing computing power resources in an intelligent computing center. Background Art

[0002] With the rapid development of technologies such as big data, artificial intelligence, and the Internet of Things, the computing demand has shown an explosive growth trend. When dealing with a large number of computing tasks, traditional computing centers have gradually exposed problems such as uneven resource allocation, high access latency, and insufficient security. To address these challenges, intelligent computing centers have emerged. An intelligent computing center is a comprehensive computing platform integrating advanced computing technologies, intelligent management systems, and efficient network facilities. It provides users with flexible and scalable computing resources by integrating multiple computing modes such as cloud computing, edge computing, and distributed computing. The core goal of an intelligent computing center is to achieve intelligent scheduling, efficient utilization, and convenient access of computing resources, so as to meet the needs of various complex computing tasks. In an intelligent computing center, the access to computing power resources is a key link connecting users and computing resources. However, the existing methods for accessing computing power resources still have many limitations.

[0003] In the Chinese invention application with the application publication number CN118842657A, a method and device for accessing computing power resources in an intelligent computing center are disclosed, including step S1: receiving an access instruction from a user, where the access instruction is used to indicate the target computing power resource to be accessed by the user, and the access instruction carries the user's access token, and the access token carries the user's identity information; step S2: performing a hash operation on the access token to obtain a to-be-verified hash value corresponding to the access token; step S3: determining whether a first condition is satisfied. If satisfied, step S4 is executed; if not satisfied, step S5 is executed; the first condition includes: the to-be-verified hash value is consistent with a reference hash value; step S4: verifying whether the user has the permission to access the target computing power resource according to the user's identity information; step S5: rejecting the user's access to the target computing power resource.

[0004] In the above invention application, by combining the hash operation and the access token, the security of identity verification and the immutability of data are ensured; by verifying the user's permissions, it is ensured that only authorized users can access specific computing power resources, thereby reducing the risks brought by unauthorized users accessing computing power resources. However, there is no restriction on the computing power access of users who already have permissions. Users may use computing power resources without restraint, resulting in an unbalanced overall load of the computing center. Some users may occupy a large amount of resources, affecting the computing tasks of other users, leading to a decline in system performance, and even causing resource contention and conflicts.

[0005] Therefore, the present invention provides a method for accessing computing power resources in an intelligent computing center. Summary of the Invention

[0006] (1) Technical problem to be solved

[0007] In view of the deficiencies of the prior art, the present invention provides a method for accessing computing power resources in an intelligent computing center. By obtaining real-time redundant computing power and real-time computing power anomaly coefficient Ycx, calculating corrected redundant computing power, and combining the computing power requirement Sx of the user's current computing power resource access task, the present invention outputs the permitted computing power Zx and permitted computing power access duration Sc for the user's current access execution. During the peak period of computing power demand, the system dynamically allocates corrected redundant computing power to ensure that the training task is completed on time, thus solving the technical problems described in the background art.

[0008] (2) Technical solution

[0009] To achieve the above object, the present invention is realized through the following technical solutions: A method for accessing computing power resources in an intelligent computing center includes the following steps:

[0010] Identify the user's identity, query the corresponding access permission according to the user's identity, obtain the single-task computing volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, and calculate the computing power requirement Sx of the user's current computing power resource access task;

[0011] Calculate the real-time redundant computing power Ry i , and obtain the maximum computing power access volume per unit Fw corresponding to the computing power resource access permission level of the user's current computing power resource access task, and send out an auxiliary access analysis warning;

[0012] After receiving the auxiliary access analysis warning, calculate the real-time computing power overage coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz of the real-time unit computing power, and calculate the real-time computing power anomaly coefficient Ycx, and send out a computing power correction instruction;

[0013] After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, calculate the corrected redundant computing power Jry i , and combine the computing power requirement Sx of the user's current computing power resource access task, and output the permitted computing power Zx and permitted computing power access duration Sc for the user's current computing power resource access for access execution.

[0014] Furthermore, identify the user's identity through an identity verification mechanism (such as username and password, digital certificate, etc.), query the corresponding access permission according to the user's identity. If the computing power resource access permission fails, exit the computing power resource access program. If the computing power resource access permission passes, obtain the user's current computing power resource access task and the computing power resource access permission level.

[0015] Among them, querying its corresponding access rights can be achieved through methods such as access control lists, role assignment tables, or attribute rule bases.

[0016] Furthermore, obtain the single-task computation volume Dj of the user's current computing power resource access task according to the PyTorch tool library, and obtain the concurrent task number Bf of the user's current computing power resource access task according to the system monitoring tool.

[0017] thop or ptflops in the PyTorch tool library: Automatically count FLOPs through the input model and example data, applicable to standard modules (such as nn.Conv2d, nn.Linear), and torch_flops supports capturing all operators in the model (including non-modular operations such as torch.add), avoiding the problem that traditional tools ignore some operations.

[0018] Among them, the calculation method of the single-task computation volume Dj is different according to the task type:

[0019] For algorithmic tasks (such as matrix multiplication), Dj = 2 × matrix dimension 3 (For example: multiplying two 1000×1000 matrices requires 2×10 9 FLOPs)

[0020] For deep learning tasks: Dj = forward FLOPs of the model + backward FLOPs (usually the backward computation volume is 2 - 3 times that of the forward)

[0021] The calculation method of the concurrent task number Bf is different according to the task type:

[0022] For real-time tasks: The concurrent task number Bf = requests per second (QPS) × single-task processing time (seconds)

[0023] For offline tasks: The concurrent task number Bf = total number of samples ÷ batch size

[0024] Furthermore, obtain the single-task computation volume Dj and the concurrent task number Bf of the user's current computing power resource access task, and calculate the computing power demand Sx of the user's current computing power resource access task:

[0025] Sx = Dj * Bf * α

[0026] Among them, α is the safety redundancy coefficient, used to reserve resources to cope with sudden loads, usually 1.2 - 2.0.

[0027] Furthermore, obtain the continuous computing power Cs and the real-time used computing power Ys of the computing power resources of the intelligent computing center according to the operation log i , and calculate the real-time redundant computing power Ry i :

[0028] Ryi = Cs - Ys i

[0029] Where i represents the time of the data.

[0030] Continuous computing power refers to the average computing power that a computing center can continuously output under the condition of long - term operation (such as several hours to several days) and full - load or near - full - load. It takes into account the limitations of the hardware in actual operation (such as heat dissipation, power supply, software scheduling, etc.), rather than relying solely on the theoretical maximum value.

[0031] Furthermore, obtain the maximum computing power access amount per unit Fw corresponding to the computing power resource access permission level of the user's current computing power resource access task. If the maximum computing power access amount per unit Fw is less than the real - time redundant computing power Ry i , then access and execute the user's current computing power resource access task according to the maximum computing power access amount per unit. If the maximum computing power access amount per unit Fw is not less than the real - time redundant computing power Ry i , then send out an auxiliary access analysis warning.

[0032] Furthermore, after receiving the auxiliary access analysis warning, obtain the computing power requirements Sx a , the computing power access amount per unit time Fw a , the computing power access duration Fs a and the completion progress Jd of the computing power access task a of each computing power resource access task in the state of being accessed and executed, and calculate the real - time computing power over - quantity coefficient δ:

[0033]

[0034] Where a represents the sequence number of each computing power resource access task in the state of being accessed and executed, a = 1, 2, …, m, and m is the total number of all computing power resource access tasks in the state of being accessed and executed. The computing power resource access task states include access request, access execution, and access completion.

[0035] Furthermore, obtain the cumulative computing power access amount Fw a , the cumulative memory occupancy Nc a and the cumulative bandwidth occupancy Dk a of each computing power resource access task in the state of being accessed and executed according to the operation log, and calculate the memory occupancy Nz and bandwidth occupancy Dz per unit real - time computing power:

[0036]

[0037] Furthermore, obtain the real - time computing power over - quantity coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz per unit real - time computing power, and calculate the real - time computing power anomaly coefficient Ycx:

[0038]

[0039] Among them, Δδ represents the standard computing power overage coefficient, ΔNz represents the memory occupancy of the standard unit computing power, and ΔDz represents the bandwidth occupancy of the standard unit computing power.

[0040] Furthermore, obtain the real-time computing power anomaly coefficient Ycx. When the real-time computing power anomaly coefficient Ycx exceeds 0.2, send out a computing power correction instruction;

[0041] When the real-time computing power anomaly coefficient Ycx does not exceed 0.2, then perform access execution on the user's current computing power resource access task according to the real-time redundant computing power Ry i for the user's current computing power resource access task.

[0042] Furthermore, after receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the corrected redundant computing power Jry i :

[0043]

[0044] Furthermore, obtain the computing power requirement Sx of the user's current computing power resource access task and the corrected redundant computing power Jry i , and calculate the permitted computing power Zx and the permitted computing power access duration Sc for the user's current computing power resource access:

[0045]

[0046] Output the permitted computing power Zx and the permitted computing power access duration Sc for the user's current computing power resource access for access execution.

[0047] (3) Beneficial effects

[0048] The present invention provides a method for accessing computing power resources in an intelligent computing center, having the following beneficial effects:

[0049] 1. Identify the user's identity, query the corresponding access rights according to the user's identity, obtain the single-task calculation amount Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, calculate the computing power requirement Sx of the user's current computing power resource access task, and can accurately allocate computing power resources, improve efficiency, reduce costs, and provide data support for strategic decision-making.

[0050] 2. Calculate the real-time redundant computing power Ry i , and obtain the maximum computing power access amount Fw per unit corresponding to the computing power resource access permission level of the user's current computing power resource access task, send out an auxiliary access analysis warning, and the system warning prevents resource exhaustion, adjusts task allocation according to the redundant computing power, and avoids node overload.

[0051] 3. After receiving the auxiliary access analysis warning, calculate the real-time computing power overage coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power, calculate the real-time computing power anomaly coefficient Ycx, send out a computing power correction instruction, monitor the computing power usage in real time, dynamically schedule tasks, and avoid node overload by adjusting task allocation to improve the overall computing efficiency.

[0052] 4. After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, calculate the corrected redundant computing power Jry i , combined with the computing power requirement Sx of the user's current computing power resource access task, output the permitted computing power Zx and permitted computing power access duration Sc for the user's current access execution. During the peak period of computing power demand, the system dynamically allocates the corrected redundant computing power to ensure the timely completion of the training task. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic flowchart of a method for accessing computing power resources of an intelligent computing center according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Please refer to Figure 1 , the present invention provides a method for accessing computing power resources of an intelligent computing center, including the following steps:

[0056] Step 1. Identify the user's identity, query the corresponding access rights according to the user's identity, obtain the single-task computing volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, and calculate the computing power requirement Sx of the user's current computing power resource access task.

[0057] The above Step 1 includes the following contents:

[0058] Step 101. Identify the user's identity through an identity authentication mechanism (such as username and password, digital certificate, etc.), query the corresponding access rights according to the user's identity. If the computing power resource access right fails, exit the computing power resource access program. If the computing power resource access right passes, obtain the user's current computing power resource access task and the computing power resource access right level.

[0059] Among them, querying its corresponding access rights can be achieved through methods such as access control lists, role assignment tables, or attribute rule bases.

[0060] Step 102: Obtain the single-task computation volume Dj of the user's current computing power resource access task according to the PyTorch tool library, and obtain the number of concurrent tasks Bf of the user's current computing power resource access task according to the system monitoring tool.

[0061] In the PyTorch tool library, thop or ptflops: Automatically count FLOPs by inputting the model and example data, applicable to standard modules (such as nn.Conv2d, nn.Linear), and torch_flops supports capturing all operators in the model (including non-modular operations such as torch.add), avoiding the problem that traditional tools ignore some operations.

[0062] Among them, the single-task computation volume Dj has different calculation methods according to different task types:

[0063] For algorithmic tasks (such as matrix multiplication), Dj = 2 × matrix dimension 3 (For example: multiplying two 1000×1000 matrices requires 2×10 9 FLOPs)

[0064] For deep learning tasks: Dj = forward FLOPs of the model + backward FLOPs (usually the backward computation volume is 2 - 3 times that of the forward)

[0065] The number of concurrent tasks Bf has different calculation methods according to different task types:

[0066] For real-time tasks: The number of concurrent tasks Bf = requests per second (QPS) × single-task processing time (seconds)

[0067] For offline tasks: The number of concurrent tasks Bf = total number of samples ÷ batch size

[0068] Step 103: Obtain the single-task computation volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, and calculate the computing power demand Sx of the user's current computing power resource access task:

[0069] Sx = Dj * Bf * α

[0070] Among them, α is the safety redundancy factor, used to reserve resources to cope with sudden loads, usually 1.2 - 2.0.

[0071] When using, combine the content in Steps 101 to 103:

[0072] Identify the user's identity, query the corresponding access rights according to the user's identity, obtain the single-task computing volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, calculate the computing power requirement Sx of the user's current computing power resource access task, and accurately allocate computing power resources, improve efficiency, reduce costs, and provide data support for strategic decision-making.

[0073] Step 2: Calculate the real-time redundant computing power Ry i and obtain the maximum unit computing power access volume Fw corresponding to the computing power resource access right level of the user's current computing power resource access task, and send out an auxiliary access analysis warning.

[0074] The said Step 2 includes the following content:

[0075] Step 201: Obtain the continuous computing power Cs and the real-time used computing power Ys of the computing power resources of the intelligent computing center according to the operation log i and calculate the real-time redundant computing power Ry i :

[0076] Ry i = Cs - Ys i

[0077] where i represents the time of the data.

[0078] The continuous computing power refers to the average computing power that the computing center can continuously output under the condition of long-term operation (such as several hours to several days) and full load or near full load. It takes into account the limitations of the hardware in actual operation (such as heat dissipation, power supply, software scheduling, etc.), rather than relying solely on the theoretical maximum value.

[0079] Step 202: Obtain the maximum unit computing power access volume Fw corresponding to the computing power resource access right level of the user's current computing power resource access task. If the maximum unit computing power access volume Fw is less than the real-time redundant computing power Ry i , then execute the access to the user's current computing power resource access task according to the maximum unit computing power access volume. If the maximum unit computing power access volume Fw is not less than the real-time redundant computing power Ry i , then send out an auxiliary access analysis warning.

[0080] When in use, combine the content in Steps 201 and 202:

[0081] Calculate the real-time redundant computing power Ry i and obtain the maximum unit computing power access volume Fw corresponding to the computing power resource access right level of the user's current computing power resource access task, and send out an auxiliary access analysis warning. The system warning prevents resource exhaustion, adjusts task allocation according to the redundant computing power, and avoids node overload.

[0082] Step 3: After receiving the auxiliary access analysis warning, calculate the real-time computing power overage coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power, calculate the real-time computing power anomaly coefficient Ycx, and send out a computing power correction instruction.

[0083] The content of Step 3 is as follows:

[0084] Step 301: After receiving the auxiliary access analysis warning, obtain the computing power requirement Sx, the computing power access volume Fw per unit time a , the computing power access duration Fs a , and the progress Jd of the computing power access task completion a of each computing power resource access task in the state of being accessed and executed, and calculate the real-time computing power overage coefficient δ: a where a represents the sequence number of each computing power resource access task in the state of being accessed and executed, a = 1, 2,..., m, and m is the total number of all computing power resource access tasks in the state of being accessed and executed. The states of the computing power resource access tasks include access request, access execution, and access completion.

[0085]

[0086]

[0087] Step 302: Obtain the cumulative computing power access volume Fw, the cumulative memory occupancy Nc a , and the cumulative bandwidth occupancy Dk a of each computing power resource access task in the state of being accessed and executed according to the operation log, and calculate the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power: a

[0088]

[0089] Step 303: Obtain the real-time computing power overage coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power, and calculate the real-time computing power anomaly coefficient Ycx:

[0090]

[0091] where Δδ represents the standard computing power overage coefficient, ΔNz represents the standard memory occupancy per unit of computing power, and ΔDz represents the standard bandwidth occupancy per unit of computing power.

[0092] Step 304: Obtain the real-time computing power anomaly coefficient Ycx. When the real-time computing power anomaly coefficient Ycx exceeds 0.2, send out a computing power correction instruction.

[0093] When the real-time computing power anomaly coefficient Ycx does not exceed 0.2, then according to the real-time redundant computing power Ry iExecute the access for the user's current computing power resource access task.

[0094] When in use, combine the content in Steps 301 to 304:

[0095] After receiving the auxiliary access analysis warning, calculate the real-time computing power overage coefficient δ, the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power, calculate the real-time computing power anomaly coefficient Ycx, send out a computing power correction instruction, monitor the computing power usage in real time, dynamically schedule tasks, and avoid node overload by adjusting task allocation to improve the overall computing efficiency.

[0096] Step Four: After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the corrected redundant computing power Jry i , combine with the computing power requirement Sx of the user's current computing power resource access task, and output the permitted computing power Zx and permitted computing power access duration Sc for the user's current computing power resource access for access execution.

[0097] The said Step Four includes the following content:

[0098] Step 401: After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the corrected redundant computing power Jry i :

[0099]

[0100] Step 402: Obtain the computing power requirement Sx of the user's current computing power resource access task and the corrected redundant computing power Jry i , and calculate the permitted computing power Zx and permitted computing power access duration Sc for the user's current computing power resource access:

[0101]

[0102] Output the permitted computing power Zx and permitted computing power access duration Sc for the user's current computing power resource access for access execution.

[0103] When in use, combine the content in Steps 401 and 402:

[0104] After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the corrected redundant computing power Jry i , combine with the computing power requirement Sx of the user's current computing power resource access task, and output the permitted computing power Zx and permitted computing power access duration Sc for the user's current computing power resource access for access execution. During the peak period of computing power demand, the system dynamically allocates the corrected redundant computing power to ensure that the training task is completed on time.

[0105] The present invention provides another specific embodiment:

[0106] Video rendering task computing power estimation: Rendering a 4K video (60fps) for 10 minutes, with ray tracing calculation required for each frame, single-frame calculation amount: 1 TFLOPs, redundancy factor: 1.5

[0107] Total calculation amount:

[0108] 10 minutes × 60 seconds / minute × 60fps × 1 TFLOPs / frame * 1.5 = 216,000 * 1.5 TFLOPS

[0109] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0110] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0111] As described above, only the specific embodiments of the present application are provided, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for accessing computing power resources of an intelligent computing center, characterized in that: Including the following steps: Identify the user's identity, query the corresponding access rights according to the user's identity, obtain the single-task computing volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, and calculate the computing power requirement Sx of the user's current computing power resource access task; Calculate the real-time redundant computing power Ry i , and obtain the maximum computing power access amount per unit Fw corresponding to the computing power resource access permission level of the user's current computing power resource access task, and send out an auxiliary access analysis warning After receiving the auxiliary access analysis warning, calculate the real-time computing power overage coefficient δ, the memory occupancy Nz and the bandwidth occupancy Dz of the real-time unit computing power, and calculate the real-time computing power anomaly coefficient Ycx, and send out a computing power correction instruction; After receiving the computing power calibration instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the calibrated redundant computing power Jry i , combined with the computing power requirement Sx of the user's current computing power resource access task, output the permitted computing power Zx and the permitted computing power access duration Sc for the user's current access execution.

2. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the single-task computing volume Dj and the number of concurrent tasks Bf of the user's current computing power resource access task, and calculate the computing power requirement Sx of the user's current computing power resource access task: Sx = Dj * Bf * α Where α is a security redundancy coefficient, which is used to reserve resources to cope with sudden loads.

3. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the continuous computing power Cs and the real-time used computing power Ys of the intelligent computing center according to the operation log i , and calculate the real-time redundant computing power Ry i : Ry i = Cs - Ys i Where i represents the time of the data.

4. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the maximum computing power access volume per unit Fw corresponding to the computing power resource access permission level of the user's current computing power resource access task. If the maximum computing power access volume per unit Fw is less than the real-time redundant computing power Ry i , then access and execute the user's current computing power resource access task according to the maximum computing power access volume per unit. If the maximum computing power access volume per unit Fw is not less than the real-time redundant computing power Ry i , then send out an auxiliary access analysis warning.

5. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: After receiving the auxiliary access analysis warning, obtain the computing power requirements Sx of each computing power resource access task in the ongoing access execution state a , the computing power access volume Fw per unit time a , the computing power access duration Fs a and the completion progress Jd of the computing power access task a , and calculate the real-time computing power overage coefficient δ: Where a represents the sequence number of each computing power resource access task in the access execution state, a = 1, 2,..., m, and m is the total number of each computing power resource access task in the access execution state. The computing power resource access task state includes access request, access execution, and access completion.

6. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the cumulative computing power access volume Fw, cumulative memory occupancy Nc, and cumulative bandwidth occupancy Dk of each computing power resource access task in the ongoing access and execution state according to the operation log, and calculate the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power: a , cumulative memory occupancy Nc a and cumulative bandwidth occupancy Dk a , and calculate the memory occupancy Nz and bandwidth occupancy Dz per unit of real-time computing power:

7. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the real-time computing power overage coefficient δ, the memory occupancy Nz and the bandwidth occupancy Dz of the real-time unit computing power, and calculate the real-time computing power anomaly coefficient Ycx: Where Δδ represents the standard computing power overage coefficient, ΔNz represents the standard memory occupancy of the unit computing power, and ΔDz represents the standard bandwidth occupancy of the unit computing power.

8. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the real-time computing power anomaly coefficient Ycx. When the real-time computing power anomaly coefficient Ycx exceeds 0.2, send out a computing power correction instruction; When the real-time computing power anomaly coefficient Ycx does not exceed 0.2, the real-time redundant computing power Ry i is used to perform the access execution for the user's current computing power resource access task.

9. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: After receiving the computing power correction instruction, obtain the real-time redundant computing power Ry i and the real-time computing power anomaly coefficient Ycx, and calculate the corrected redundant computing power Jry i :

10. The computing power resource access method of an intelligent computing center according to claim 1, characterized in that: Obtain the computing power requirement Sx and the corrected redundant computing power Jry of the user's current computing power resource access task i , and calculate the allowed computing power Zx and the allowed computing power access duration Sc of the user's current computing power resource access: Output the permitted computing power Zx and the permitted computing power access duration Sc of the user's current computing power resource access for access execution.

Images