Multi-cell task unloading algorithm in edge computing environment

Abstract

The invention relates to the field of computer algorithms, in particular to a multi-cell task unloading algorithm in an edge computing environment, which comprises the following steps of: dividing a network into a plurality of cell environments, determining a main environment, issuing computing tasks to the main environment and each cell environment by utilizing a cloud platform, acquiring response time of each cell environment by a responder, taking the response time as a training sample, and outputting the training sample; the method comprises the following steps: dividing into a plurality of sub-samples according to a preset learning rate, training a plurality of response learning models by using the plurality of sub-samples, learning response time, adjusting a main environment and a plurality of cell environments according to a learning result, and obtaining a response time threshold value by setting a response time threshold value and the response learning models according to the response time. According to the method, whether the plurality of cell environments need to be adjusted or not is determined, and the response learning model is utilized to visualize the response efficiency, so that the operation efficiency of the multi-cell environment is improved while the accuracy of response time identification among the cell environments is effectively improved, and the complexity of edge computing environment management is reduced.

Description

Technical Field

[0001] The present invention relates to the field of computer algorithms, and in particular to a multi-cell task offloading algorithm in an edge computing environment. Background Art

[0002] With the rapid development of technologies such as the Internet of Things and artificial intelligence, edge computing has gradually become a research hotspot. In edge computing, task offloading is a key issue, which involves multiple aspects such as energy consumption, computing resource utilization, and response time. The multi-cell task offloading algorithm aims to solve the problem of how to efficiently transfer computing tasks from devices or terminals to edge servers for computing in a multi-cell environment.

[0003] Chinese patent application publication number: CN113296845A discloses a multi-cell task offloading algorithm based on deep reinforcement learning in an edge computing environment, proposes an improved dual-depth Q network algorithm, and designs a task offloading algorithm based on the dual-depth Q network algorithm, and uses the MEC edge control platform as an intelligent agent to interact with the Internet of Vehicles environment, aiming to optimize the task processing rate and task execution delay under the same resources. In addition, in complex business scenarios, the Q value function in the DQN algorithm is improved to improve the model's ability to select the optimal offloading strategy in response to the over-estimation problem in the DQN algorithm. This solution can significantly improve the utilization of computing and storage resources, effectively reduce the task execution delay of vehicle terminals, and greatly improve the utilization of edge-side computing and storage resources. The algorithm convergence efficiency is higher than that of the traditional DQN algorithm, which solves the problem of the optimal offloading ratio of tasks and can more effectively reduce the execution delay of tasks.

[0004] China Patent Authorization Announcement No.: CN111586720B discloses a joint optimization method for task offloading and resource allocation in a multi-cell scenario, which belongs to the field of mobile edge computing. The method includes: first, establishing an MEC task offloading model in a multi-cell scenario and designing a system overhead function. Then, a chaotic mutation binary particle swarm algorithm is used to optimize the user's offloading decision; when the user's offloading decision is obtained, the original problem is decomposed into two sub-problems: MEC computing resource allocation and uplink sub-channel allocation; the Lagrange multiplier method is used to allocate MEC computing resources to the offloading user, and the improved Kuhn-Munkres algorithm is used to allocate uplink sub-channels to the offloading user under the constraints of the user's minimum rate and maximum tolerable interference; the invention can reduce the system overhead of users performing tasks and effectively improve the performance of the system.

[0005] However, the above method has the following problems: it is impossible to perform cluster control on a multi-cell environment in real time, and the algorithm operation efficiency is low. Summary of the invention

[0006] To this end, the present invention provides a multi-cell task offloading algorithm in an edge computing environment to overcome the problem in the prior art that the multi-cell environment cannot be clustered and controlled in real time and the algorithm operation efficiency is low.

[0007] To achieve the above object, the present invention provides a multi-cell task offloading algorithm in an edge computing environment, comprising:

[0008] Dividing the network into a plurality of cell environments, and determining a main environment among the plurality of cell environments;

[0009] Utilize the cloud platform to issue computing tasks to the main environment and each cell environment;

[0010] The responder obtains the response time of each cell environment in response to the computing task;

[0011] Taking the response time as a training sample, dividing it into a plurality of sub-samples according to a preset learning rate;

[0012] Using the sub-samples to train a plurality of response learning models respectively, and learning the response time;

[0013] Adjusting the main environment and the plurality of cell environments according to the learning result;

[0014] Wherein, the computing tasks are divided into primary tasks and secondary tasks according to their priorities;

[0015] The response learning model is provided with a response efficiency threshold;

[0016] The preset learning rate is related to the minimum recognition capability of the response learning model.

[0017] Further, the network is divided into the main environment and a plurality of cell environments, wherein:

[0018] For the main environment, it is provided with a main processor and a receiving cloud for responding to the main task and receiving the response time of the cell environment;

[0019] The cell environment is connected to the main environment and is provided with a secondary processor and a transmitter for responding to the secondary task and sending the response time to the main environment.

[0020] Furthermore, after receiving the online message from the receiving cloud, the main processor determines the cell environment where the receiving cloud is located as the main environment.

[0021] Furthermore, the cloud platform is provided with a release time, and the computing task is released to the main environment and each cell environment at every release time.

[0022] Wherein, a timer is provided in the cloud platform for controlling and calculating the publishing duration.

[0023] Furthermore, the responder is connected to the main processor to obtain the response time of the main environment and the cell environment.

[0024] The response time is the time taken for the main environment and the cell environment to receive the computing task and complete the computing task.

[0025] Furthermore, the response time is marked as a training sample, and when dividing into a plurality of sub-samples, the plurality of response times are sorted according to the preset learning rate and correspond to each sub-sample.

[0026] Furthermore, the step of training the response learning model includes:

[0027] The sub-sample iteratively optimizes the response learning model until the loss value converges to a preset value;

[0028] Testing of current response learning models;

[0029] When the test result meets the preset test result, the parameters of the response learning model are saved.

[0030] Further, the corresponding indicator features are generated according to the sub-samples, the sub-samples are fitted, and a number of processed data are generated, the obtained number of processed data are divided according to the preset learning rate, corresponding pre-processed data are generated, and the pre-processed data are passed through the response learning model to obtain a corresponding response efficiency distribution diagram,

[0031] The fitting is to clean the subsamples based on the indicator characteristics.

[0032] The indicator characteristics include the duration of the response time, the preset learning rate size and / or the priority of the computing task.

[0033] Further, the response efficiency distribution graph is compared with the response efficiency threshold, and when the response efficiency distribution graph is greater than the response efficiency threshold, the main environment and the plurality of cell environments are not adjusted.

[0034] Furthermore, when the response efficiency distribution graph is less than the response efficiency threshold, part of the cell environments are re-marked as primary environments.

[0035] Compared with the prior art, the present invention divides the network into several cell environments and determines the main environment, uses the cloud platform to publish computing tasks to the main environment and each cell environment, the responder obtains the response time of each cell environment, and uses it as a training sample, divides it into several sub-samples according to a preset learning rate, uses several sub-samples to train several response learning models respectively, learns the response time, adjusts the main environment and several cell environments according to the learning results, sets the response time threshold and the response learning model, determines whether several cell environments need to be adjusted according to the response time, and uses the response learning model to visualize the response efficiency, which effectively improves the accuracy of response time identification between cell environments while improving the operation efficiency of multi-cell environments, thereby reducing the complexity of edge computing environment management.

[0036] Furthermore, by setting up a main environment in several cell environments and setting up a receiving cloud in the main environment, the response time of several cell environments can be collected and updated through the main environment, thereby improving the operation efficiency of the multi-cell environment.

[0037] Furthermore, by setting the release time in the cloud platform, computing tasks are released to the main environment and each cell environment at every release time, which avoids the error in response time caused by different release times, improves the accuracy of response time collection, and thus improves the accuracy of machine learning results.

[0038] Furthermore, by sorting several response times according to a preset learning rate and corresponding them to each sub-sample when dividing them into several sub-samples, the response time data can be updated in real time, which effectively improves the accuracy of response time identification between cell environments and improves the operating efficiency of multi-cell environments.

[0039] Furthermore, by segmenting the obtained processed data according to a preset learning rate to generate corresponding preprocessed data, the recognition efficiency of the response time is improved, so that effective features of the response time can be quickly selected, thereby further improving the accuracy of the machine learning results.

[0040] Furthermore, by setting a response efficiency threshold in the response learning model, it is possible to quickly distinguish whether adjustments need to be made to several cell environments, and the response efficiency is visualized using the response learning model, thereby reducing the complexity of edge computing environment management. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the multi-cell task offloading algorithm in the edge computing environment of the present invention;

[0042] Figure 2 Flowchart for training a response learning model for the present invention. DETAILED DESCRIPTION

[0043] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0045] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0046] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] See also Figure 1 As shown, it is a flow chart of a multi-cell task offloading algorithm in an edge computing environment of the present invention, including:

[0048] Step S1, dividing the network into a plurality of cell environments, and determining a main environment among the plurality of cell environments;

[0049] Step S2, using the cloud platform to publish computing tasks to the main environment and each community environment;

[0050] Step S3, the responder obtains the response time of each cell environment response calculation task;

[0051] Step S4, taking the response time as a training sample and dividing it into a number of sub-samples according to a preset learning rate;

[0052] Step S5, using the plurality of sub-samples to respectively train a plurality of response learning models, and learning the response time;

[0053] Step S6, adjusting the main environment and several cell environments according to the learning result;

[0054] Among them, computing tasks are divided into primary tasks and secondary tasks according to priority;

[0055] In the specific implementation, the division of main tasks and secondary tasks is related to the duration of response time, the size of preset learning rate and / or the priority of computing tasks. For a computing environment with no more than 3 parallel computing tasks, the main tasks and secondary tasks can be randomly divided. For a computing environment with more than 3 parallel computing tasks, the priorities of main tasks and secondary tasks are divided sequentially according to the priority of computing tasks, the size of preset learning rate and the duration of response time.

[0056] A response efficiency threshold is set in the response learning model;

[0057] The preset learning rate is related to the minimum recognition ability of the response learning model.

[0058] In a specific implementation, when the preset learning rate is set to 0.1, the response learning model has the best learning effect on response efficiency.

[0059] By dividing the network into several cell environments and determining the main environment, the cloud platform is used to publish computing tasks to the main environment and each cell environment. The responder obtains the response time of each cell environment and uses it as a training sample. It is divided into several sub-samples according to a preset learning rate, and several response learning models are trained using the sub-samples to learn the response time. The main environment and several cell environments are adjusted according to the learning results. By setting the response time threshold and the response learning model, it is determined whether several cell environments need to be adjusted according to the response time, and the response efficiency is visualized using the response learning model. While effectively improving the accuracy of response time identification between cell environments, the operation efficiency of multi-cell environments is improved, thereby reducing the complexity of edge computing environment management.

[0060] Specifically, the network is divided into a main environment and several cell environments, among which:

[0061] For the main environment, it has a main processor and a receiving cloud to respond to the main task and receive the response time of the cell environment;

[0062] The cell environment is connected to the main environment and is provided with a secondary processor and a transmitter for responding to the secondary task and sending the response time to the main environment.

[0063] In a specific implementation, the main processor is used to process computing tasks with smaller capacity and more complex calculations, and the secondary processor is used to process computing tasks with larger capacity and simpler calculations.

[0064] By setting up a main environment in several cell environments and setting up a receiving cloud in the main environment, the response time of several cell environments can be collected and updated through the main environment, thereby improving the operation efficiency of the multi-cell environment.

[0065] Specifically, after receiving the online message from the receiving cloud, the main processor determines the cell environment where the receiving cloud is located as the main environment.

[0066] Specifically, the cloud platform has a release period, and computing tasks are released to the main environment and each community environment at each release period.

[0067] Among them, a timer is set in the cloud platform to control and calculate the publishing time.

[0068] Embodiment 1:

[0069] If the release duration is set to 0.1s, the timer in the cloud platform will release computing tasks to the main environment and each community environment at 0.1s, 0.2s, 0.3s, 0.4s, 0.5s...

[0070] By setting the release time in the cloud platform, computing tasks are released to the main environment and each community environment at every release time, which avoids the error in response time caused by different release times, improves the accuracy of response time collection, and thus improves the accuracy of machine learning results.

[0071] Specifically, the responder is connected to the main processor to obtain the response time of the main environment and the cell environment.

[0072] The response time is the time from when the main environment and the cell environment receive the computing task to when the computing task is completed.

[0073] Specifically, the response time is marked as a training sample, and when dividing into a plurality of sub-samples, the plurality of response times are sorted according to a preset learning rate and correspond to each sub-sample.

[0074] Embodiment 2:

[0075] If the response time of A is 0.1s, the response time of B is 0.3s, the response time of C is 0.2s, and the preset learning rate is 0.1, then the order of response time is A response time < B response time < C response time.

[0076] By sorting several response times according to the preset learning rate when dividing several sub-samples and corresponding them to each sub-sample, the response time data can be updated in real time, which effectively improves the accuracy of response time identification between cell environments and improves the operation efficiency of multi-cell environments.

[0077] See also Figure 2 As shown, it is a flow chart of the present invention for training the response learning model, including:

[0078] Step St1, the sub-sample response learning model is iteratively optimized until the loss value converges to a preset value;

[0079] Step St2, testing the current response learning model;

[0080] Step St3, when the test result meets the preset test result, the parameters of the response learning model are saved.

[0081] Specifically, the corresponding indicator features are generated according to the sub-samples, the sub-samples are fitted, and a number of processed data are generated. The obtained processed data are divided according to the preset learning rate to generate corresponding pre-processed data. The pre-processed data is passed through the response learning model to obtain the corresponding response efficiency distribution diagram.

[0082] Among them, fitting is to clean the sub-samples based on the indicator characteristics.

[0083] Indicator characteristics include the duration of the response time, the preset learning rate size and / or the priority of the computing task.

[0084] Embodiment 3:

[0085] Taking Task A, Task B, Task C, and Task D as examples, the priority of the computing tasks is recorded as 1, 2, 3, and 4, where 1 is the highest priority:

[0086] Task A: The computing task priority is level 1, the preset learning rate is 0.01, and the response time is 2s;

[0087] Task B: The computing task priority is level 2, the preset learning rate is 0.01, and the response time is 5s;

[0088] Task C: The computing task priority is level 3, the preset learning rate is 0.01, and the response time is 3s;

[0089] Task D: The computing task priority is level 4, the preset learning rate is 0.01, and the response time is 10s;

[0090] At this time, task A is taken as the main task. When task A is completed in the 2nd second, task B is carried out as the main task.

[0091] Embodiment 4:

[0092] Taking Task A, Task B, Task C, and Task D as examples, the priority of the computing tasks is recorded as 1, 2, 3, and 4, where 1 is the highest priority:

[0093] Task A: The computing task priority is level 1, the preset learning rate is 0.01, and the response time is 2s;

[0094] Task B: The computing task priority is level 2, the preset learning rate is 0.01, and the response time is 5s;

[0095] Task C: The computing task priority is level 1, the preset learning rate is 0.01, and the response time is 3s;

[0096] Task D: The computing task priority is level 2, the preset learning rate is 0.01, and the response time is 10s;

[0097] At this time, task A is taken as the main task. When task A is completed in the 2nd second, task C is carried out as the main task.

[0098] Embodiment 5:

[0099] Taking Task A, Task B, Task C, and Task D as examples, the priority of the computing tasks is recorded as 1, 2, 3, and 4, where 1 is the highest priority:

[0100] Task A: The computing task priority is level 1, the preset learning rate is 0.01, and the response time is 2s;

[0101] Task B: The computing task priority is level 1, the preset learning rate is 0.01, and the response time is 5s;

[0102] Task C: The computing task priority is level 1, the preset learning rate is 0.02, and the response time is 3s;

[0103] Task D: The computing task priority is level 1, the preset learning rate is 0.02, and the response time is 10s;

[0104] At this time, task A is taken as the main task. When task A is completed in the 2nd second, task B is carried out as the main task.

[0105] By dividing the obtained processed data according to a preset learning rate and generating corresponding preprocessed data, the recognition efficiency of the response time is improved, so that the effective features of the response time can be quickly selected, thereby further improving the accuracy of the machine learning results.

[0106] Specifically, the response efficiency distribution graph is compared with the response efficiency threshold, and when the response efficiency distribution graph is greater than the response efficiency threshold, the main environment and the plurality of cell environments are not adjusted.

[0107] Embodiment 6:

[0108] The response efficiency threshold is set to 90%. If the response efficiency in the response efficiency distribution diagram is 95%, 95% is greater than the response efficiency threshold, and the main environment and several cell environments are not adjusted.

[0109] Specifically, when the response efficiency distribution graph is less than the response efficiency threshold, some cell environments are re-marked as primary environments.

[0110] Embodiment 7:

[0111] Set the response efficiency threshold to 90%. If the response efficiency in the response efficiency distribution diagram is 85%, 85% is less than the response efficiency threshold. Adjust the main environment and several cell environments, reorder the priorities of computing tasks, obtain new priorities based on the response learning model, upgrade some secondary processors to the main processor, and remark the cell environment where the upgraded secondary processors are located as the main environment.

[0112] By setting a response efficiency threshold in the response learning model, it is possible to quickly distinguish whether adjustments need to be made to several cell environments, and the response efficiency is visualized using the response learning model, thereby reducing the complexity of edge computing environment management.

[0113] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-cell task offloading algorithm in an edge computing environment, characterized in that: include: Dividing the network into a plurality of cell environments, and determining a main environment among the plurality of cell environments; Utilize the cloud platform to issue computing tasks to the main environment and each cell environment; The responder obtains the response time of each cell environment in response to the computing task; Taking the response time as a training sample, dividing it into a plurality of sub-samples according to a preset learning rate; Using the plurality of sub-samples to respectively train a plurality of response learning models, and learning the response time; Adjusting the main environment and the plurality of cell environments according to the learning result; Wherein, the computing tasks are divided into primary tasks and secondary tasks according to their priorities; The response learning model is provided with a response efficiency threshold; The preset learning rate is related to the minimum recognition capability of the response learning model.

2. According to the multi-cell task offloading algorithm in the edge computing environment of claim 1, it is characterized in that: The network is divided into the main environment and a plurality of cell environments, wherein: For the main environment, it is provided with a main processor and a receiving cloud for responding to the main task and receiving the response time of the cell environment; The cell environment is connected to the main environment and is provided with a secondary processor and a transmitter for responding to the secondary task and sending the response time to the main environment.

3. The multi-cell task offloading algorithm in the edge computing environment according to claim 2 is characterized in that: After receiving the online message from the receiving cloud, the main processor determines the cell environment where the receiving cloud is located as the main environment.

4. The multi-cell task offloading algorithm in the edge computing environment according to claim 3 is characterized in that: The cloud platform is provided with a release time, and the computing task is released to the main environment and each cell environment at every release time. Wherein, a timer is provided in the cloud platform for controlling and calculating the publishing duration.

5. The multi-cell task offloading algorithm in the edge computing environment according to claim 4 is characterized in that: The responder is connected to the main processor and acquires the response time of the main environment and the cell environment. The response time is the time taken for the main environment and the cell environment to receive the computing task and complete the computing task.

6. The multi-cell task offloading algorithm in the edge computing environment according to claim 5 is characterized in that: The response time is marked as a training sample, and when dividing into a plurality of sub-samples, the plurality of response times are sorted according to the preset learning rate and corresponded to each sub-sample.

7. The multi-cell task offloading algorithm in the edge computing environment according to claim 6 is characterized in that: The step of training the response learning model comprises: The sub-sample iteratively optimizes the response learning model until the loss value converges to a preset value; Testing of current response learning models; When the test result meets the preset test result, the parameters of the response learning model are saved.

8. The multi-cell task offloading algorithm in the edge computing environment according to claim 7 is characterized in that: Generate corresponding indicator features according to the sub-samples, fit the sub-samples, and generate a number of processed data, divide the obtained processed data according to the preset learning rate, generate corresponding pre-processed data, pass the pre-processed data through the response learning model, and obtain a corresponding response efficiency distribution diagram, The fitting is to clean the subsamples based on the indicator characteristics. The indicator characteristics include the duration of the response time, the preset learning rate size and / or the priority of the computing task.

9. The multi-cell task offloading algorithm in the edge computing environment according to claim 8 is characterized in that: The response efficiency distribution graph is compared with the response efficiency threshold, and when the response efficiency distribution graph is greater than the response efficiency threshold, the main environment and the plurality of cell environments are not adjusted.

10. The multi-cell task offloading algorithm in the edge computing environment according to claim 9 is characterized in that: When the response efficiency distribution graph is less than the response efficiency threshold, part of the cell environments are re-marked as main environments.

Images