An edge computing load balancing method for power distribution network

Abstract

The present invention discloses a method for load balancing of edge computing in a distribution network. The method continuously monitors the load-performance ratio of edge computing devices, identifies overloaded devices, and performs task migration. A method for load balancing of edge computing in a distribution network is to select the optimal target device for task migration using a graph model search algorithm, and generate a task migration plan in combination with a priority-based heuristic algorithm. The method also introduces a deep reinforcement learning algorithm, and continuously optimizes the task allocation strategy with the load balancing effect of the entire network, the quality of task execution, and the network bandwidth utilization as optimization goals. The present invention effectively solves the problems of unbalanced load of edge computing devices and limited network bandwidth resources in the distribution network, realizes the efficient utilization of computing resources and the improvement of task execution efficiency, and provides strong support for the intelligent operation and maintenance of the distribution network.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to an edge computing load balancing method for a power distribution network. Background Art

[0002] With the development of smart grids, a large number of edge computing devices are deployed in distribution networks for real-time data processing and control decisions. However, due to factors such as the wide geographical distribution of distribution networks, complex network topology, and dynamic changes in load, the computing resource utilization of edge computing devices is uneven. Some devices may be overloaded while others are idle, which not only reduces the overall computing efficiency but also may affect the stable operation of the distribution network. At the same time, the bandwidth resources of the distribution network are limited, and large amounts of data transmission may cause network congestion, affecting the execution of tasks with high real-time requirements. How to achieve dynamic load balancing between edge computing devices under limited network bandwidth conditions has become a key issue that needs to be solved urgently.

[0003] In addition, the priorities and execution quality requirements of different tasks are different. How to achieve overall load balance of the entire network while ensuring the execution quality of high-priority tasks is also a thorny problem. The solution to these problems is not only related to the operating efficiency and reliability of the distribution network, but also directly affects the overall performance and user experience of the smart grid. Therefore, how to design an efficient, flexible and scalable load balancing method to adapt to the complex and changeable operating environment of the distribution network has become a technical problem that urgently needs to be broken through. Summary of the invention

[0004] In order to make up for the deficiencies of the prior art, the present invention proposes an edge computing load balancing method for a power distribution network.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] An edge computing load balancing method for a power distribution network, characterized by comprising:

[0007] S1: Obtain topological structure data of the power distribution network and establish a graph model of the topological structure data, where nodes in the graph model represent edge computing devices and edges represent connection relationships between devices;

[0008] S2: Obtain the current load data, device performance parameters and network bandwidth information of the edge computing device, wherein the load data includes the CPU utilization, memory occupancy and storage space utilization, and calculate the load-performance ratio of each edge computing device;

[0009] S3: Obtain a capacity threshold of the edge computing device, compare the load-performance ratio with the capacity threshold, and if the load-performance ratio exceeds the capacity threshold, determine the edge computing device as an overloaded device, and obtain an overloaded device set and load data of the overloaded device set based on the overloaded device;

[0010] S4: for each device in the overloaded device set, search the neighboring devices connected to each device through the graph model, obtain the load performance ratio and computing storage resource information of the neighboring devices, and select the device with the best load performance ratio and computing storage resource information that meets the preset threshold as the task migration target device according to the load performance ratio and computing storage resource information, and obtain the migration task;

[0011] S5: Based on the load data of the overloaded device set, a task classification algorithm is used to divide the migration tasks into migratable tasks and non-migratable tasks;

[0012] S6: for the migratable tasks, a heuristic algorithm is used to generate a task migration plan, a global task allocation table of the overloaded device is obtained, and the global task allocation table is updated according to the task migration plan to obtain an updated global task allocation table; the non-migratable tasks are retained on the current overloaded device;

[0013] S7: According to the updated global task allocation table, migrate the migratable tasks to the task migration target device through the task migration interface;

[0014] S8: Continuously monitor the load-performance ratio of edge computing devices. If the load-performance ratio exceeds the capacity threshold of the edge computing device again, re-execute S2 to S7 based on the latest load data, device performance parameters, and network bandwidth information, wherein the load data includes CPU utilization, memory occupancy, and storage space utilization, until the load-performance ratios of all edge computing devices are within the capacity threshold of the edge computing devices.

[0015] Preferably, the method further includes S9, using a deep reinforcement learning algorithm to design a reward function, wherein the calculation formula of the reward function is:

[0016] (1)

[0017] in, represents the reward function, , and are the corresponding weight coefficients, Indicates the load balancing effect of the entire network. Indicates the quality of task execution, Indicates network bandwidth utilization;

[0018] The optimal task allocation strategy is learned through the reward function. The learned task allocation strategy is used to dynamically adjust the task allocation according to the real-time edge computing device load performance ratio, task execution quality and network bandwidth utilization. If a new edge computing device is connected to the power distribution network or the device performance parameters change, the updated network topology data and device data are obtained, and S1 to S8 are re-executed.

[0019] Preferably, the graph database Neo4j is used in S1 to establish a graph model of topological structure data.

[0020] Preferably, in S2, current load data, device performance parameters and network bandwidth information of each edge computing device are obtained through a network monitoring tool.

[0021] Preferably, the load performance ratio calculation formula in S2 is:

[0022] (2)

[0023] in, Indicates the load performance ratio, , , represents the weight coefficient, Indicates the CPU utilization, Indicates the memory usage. Indicates the storage space usage. Indicates device performance parameters.

[0024] Preferably, in S4, a graph search algorithm of Neo4j is used to search for neighboring devices connected to each device in the overloaded device set.

[0025] Preferably, in S5, the load data of the overloaded device set is used to divide the migration tasks into migratable tasks and non-migratable tasks using the K-means clustering algorithm. Tasks whose preset resource occupancy exceeds a preset threshold or whose set response time requirement exceeds a preset threshold are marked as non-migratable tasks, and the remaining tasks are marked as migratable tasks.

[0026] Preferably, in S6, a greedy algorithm is used to generate a task migration plan for the migratable tasks, and each time a migratable task with the highest load-performance ratio and the smallest resource requirement is selected and assigned to a migration target device with the best load-performance ratio and large resources, and the global task allocation table is updated.

[0027] Preferably, the global task allocation table updated in S7 migrates the migratable tasks to the task migration target device through the task migration interface of Kubernetes.

[0028] Preferably, the Q-Learning reinforcement learning algorithm is used in S9 to optimize the load balancing effect of the entire network, the task execution quality and the network bandwidth utilization.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The present invention discloses a method for load balancing of edge computing in a power distribution network, which method continuously monitors the load-performance ratio of edge computing devices, identifies overloaded devices and performs task migration. A method for load balancing of edge computing in a power distribution network is to select the optimal target device for task migration using a graph model search algorithm, and generate a task migration plan in combination with a priority-based heuristic algorithm. The method also introduces a deep reinforcement learning algorithm, which continuously optimizes the task allocation strategy with the load balancing effect of the entire network, the quality of task execution and the utilization rate of network bandwidth as optimization goals. The present invention effectively solves the problems of unbalanced load of edge computing devices and limited network bandwidth resources in the power distribution network, realizes the efficient utilization of computing resources and the improvement of task execution efficiency, and provides strong support for the intelligent operation and maintenance of the power distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flow chart of an edge computing load balancing method for a power distribution network. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Example 1

[0034] like Figure 1 As shown, a method for edge computing load balancing of a power distribution network includes:

[0035] S1: Obtain topological structure data of the power distribution network and establish a graph model of the topological structure data, where nodes in the graph model represent edge computing devices and edges represent connection relationships between devices;

[0036] S2: Obtain the current load data, device performance parameters and network bandwidth information of the edge computing device, wherein the load data includes the CPU utilization, memory occupancy and storage space utilization, and calculate the load-performance ratio of each edge computing device;

[0037] S3: Obtain a capacity threshold of the edge computing device, compare the load-performance ratio with the capacity threshold, and if the load-performance ratio exceeds the capacity threshold, determine the edge computing device as an overloaded device, and obtain an overloaded device set and load data of the overloaded device set based on the overloaded device;

[0038] S4: for each device in the overloaded device set, search the neighboring devices connected to each device through the graph model, obtain the load performance ratio and computing storage resource information of the neighboring devices, and select the device with the best load performance ratio and computing storage resource information that meets the preset threshold as the task migration target device according to the load performance ratio and computing storage resource information, and obtain the migration task;

[0039] S5: Based on the load data of the overloaded device set, a task classification algorithm is used to divide the migration tasks into migratable tasks and non-migratable tasks;

[0040] S6: for the migratable tasks, a heuristic algorithm is used to generate a task migration plan, a global task allocation table of the overloaded device is obtained, and the global task allocation table is updated according to the task migration plan to obtain an updated global task allocation table; the non-migratable tasks are retained on the current overloaded device;

[0041] S7: According to the updated global task allocation table, migrate the migratable tasks to the task migration target device through the task migration interface;

[0042] S8: Continuously monitor the load-performance ratio of edge computing devices. If the load-performance ratio exceeds the capacity threshold of the edge computing device again, re-execute S2 to S7 based on the latest load data, device performance parameters, and network bandwidth information, wherein the load data includes CPU utilization, memory occupancy, and storage space utilization, until the load-performance ratios of all edge computing devices are within the capacity threshold of the edge computing device.

[0043] Preferably, the method further includes S9, using a deep reinforcement learning algorithm to design a reward function, wherein the calculation formula of the reward function is:

[0044] (1)

[0045] in, represents the reward function, , and are the corresponding weight coefficients, Indicates the load balancing effect of the entire network. Indicates the quality of task execution, Indicates network bandwidth utilization;

[0046] The optimal task allocation strategy is learned through the reward function. The learned task allocation strategy is used to dynamically adjust the task allocation according to the real-time edge computing device load performance ratio, task execution quality and network bandwidth utilization. If a new edge computing device is connected to the power distribution network or the device performance parameters change, the updated network topology data and device data are obtained, and S1 to S8 are re-executed.

[0047] The distribution network topology data is the basis for edge computing task scheduling.

[0048] By establishing a graph model, the connection relationship between edge computing devices can be intuitively represented.

[0049] The load-performance ratio can reflect the actual working efficiency of the equipment.

[0050] Example 2

[0051] Preferably, the graph database Neo4j is used in S1 to establish a graph model of topological structure data.

[0052] Preferably, in S2, current load data, device performance parameters and network bandwidth information of each edge computing device are obtained through a network monitoring tool.

[0053] Preferably, the load performance ratio calculation formula in S2 is:

[0054] (2)

[0055] in, Indicates the load performance ratio, , , represents the weight coefficient, Indicates the CPU utilization, Indicates the memory usage. Indicates the storage space usage. Indicates device performance parameters.

[0056] Preferably, in S3, the capacity threshold of the edge computing device is set to 0.8, and by comparing the load performance ratio of each device with the threshold, it is determined whether the edge computing device is overloaded. If the load performance ratio exceeds 0.8, the device is identified as an overloaded device and added to the overloaded device set.

[0057] Preferably, in S4, a graph search algorithm of Neo4j is used to search for neighboring devices connected to each device in the overloaded device set.

[0058] If there is no neighboring device that meets the conditions, the search range is expanded to neighboring devices of the neighboring device until a suitable migration target device is found.

[0059] For each device in the overloaded device set, use Neo4j's graph search algorithm to search all its one-hop neighboring device nodes to obtain the load-performance ratio and currently available CPU, memory, and storage resource information of these neighboring device nodes. Sort the neighboring devices from low to high according to their load-performance ratio, and consider the resource availability of the device, and select the device with the lowest load-performance ratio and resources that meet the requirements of the migration task as the migration target device.

[0060] Preferably, the load data of the overloaded device set in S5 is used to divide the migration tasks into migratable tasks and non-migratable tasks using the K-means clustering algorithm. Tasks with a high threshold for preset resource occupancy and a high threshold for set response time are marked as non-migratable tasks, and the remaining tasks are marked as migratable tasks.

[0061] The calculation formula of K-means clustering algorithm is:

[0062] (3)

[0063] represents the clustering objective function, represents the number of clusters, represents the number of samples, Indicates Class samples, Indicates The cluster center of the class. This formula is used in the K-means clustering algorithm to divide tasks into transferable tasks and non-transferable tasks.

[0064] Preferably, in S6, a greedy algorithm is used to generate a task migration plan for the migratable tasks, and each time a migratable task with the highest load-performance ratio and the smallest resource requirement is selected and assigned to a migration target device with the best load-performance ratio and large resources, and the global task allocation table is updated.

[0065] This process is repeated until all migratable tasks are reallocated.

[0066] Task migration is performed on the overloaded device and the task migration target device respectively, and the task allocation table and load-performance ratio of the two are updated.

[0067] The global task allocation table contains information such as task ID, task load, original deployment device, and migration target device.

[0068] Preferably, the global task allocation table updated in S7 migrates the migratable tasks to the task migration target device through the task migration interface of Kubernetes.

[0069] That is, through the Kubernetes task migration interface, the task image and data are migrated and deployed from the original device to the target device, and the task routing configuration is updated.

[0070] Preferably, in S4, a device with the lowest load-performance ratio and resources that meet the requirements of the migration task is selected as the migration target device.

[0071] Set to collect equipment load data every 5 minutes for monitoring.

[0072] The load-performance ratio of the edge computing devices is continuously monitored. If the load-performance ratio exceeds 0.8 again, the above S2 to S7 are re-executed until the load-performance ratios of all devices are below 0.8 to achieve dynamic balancing of the edge computing load.

[0073] Preferably, the Q-Learning reinforcement learning algorithm is used in S9 to optimize the load balancing effect of the entire network, the task execution quality and the network bandwidth utilization.

[0074] The load balancing of the entire network and the quality of task execution before and after each task migration are recorded, and a reinforcement learning model is constructed. With the load balancing effect of the entire network, the quality of task execution, and the network bandwidth utilization as the optimization goals, the agent is guided to learn the optimal task allocation strategy through the reward function, and the task allocation strategy trained by the reinforcement learning model is applied to the actual edge computing load balancing method, and the task allocation is dynamically adjusted to achieve load balancing optimization of the edge computing network.

[0075] The present invention discloses a method for load balancing of edge computing in a power distribution network, which method continuously monitors the load-performance ratio of edge computing devices, identifies overloaded devices and performs task migration. A method for load balancing of edge computing in a power distribution network is to select the optimal target device for task migration using a graph model search algorithm, and generate a task migration plan in combination with a priority-based heuristic algorithm. The method also introduces a deep reinforcement learning algorithm, which continuously optimizes the task allocation strategy with the load balancing effect of the entire network, the quality of task execution and the utilization rate of network bandwidth as optimization goals. The present invention effectively solves the problems of unbalanced load of edge computing devices and limited network bandwidth resources in the power distribution network, realizes the efficient utilization of computing resources and the improvement of task execution efficiency, and provides strong support for the intelligent operation and maintenance of the power distribution network.

[0076] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An edge computing load balancing method for a power distribution network, characterized in that: include: S1: Obtain topological structure data of the power distribution network and establish a graph model of the topological structure data, where nodes in the graph model represent edge computing devices and edges represent connection relationships between devices; S2: Obtain the current load data, device performance parameters and network bandwidth information of the edge computing device, wherein the load data includes the CPU utilization, memory occupancy and storage space utilization, and calculate the load-performance ratio of each edge computing device; S3: Obtain a capacity threshold of the edge computing device, compare the load-performance ratio with the capacity threshold, and if the load-performance ratio exceeds the capacity threshold, determine the edge computing device as an overloaded device, and obtain an overloaded device set and load data of the overloaded device set based on the overloaded device; S4: for each device in the overloaded device set, search the neighboring devices connected to each device through the graph model, obtain the load performance ratio and computing storage resource information of the neighboring devices, and select the device with the best load performance ratio and computing storage resource information that meets the preset threshold as the task migration target device according to the load performance ratio and computing storage resource information, and obtain the migration task; S5: Based on the load data of the overloaded device set, a task classification algorithm is used to divide the migration tasks into migratable tasks and non-migratable tasks; S6: for the migratable tasks, a heuristic algorithm is used to generate a task migration plan, a global task allocation table of the overloaded device is obtained, and the global task allocation table is updated according to the task migration plan to obtain an updated global task allocation table; Keep non-migratable tasks on the current overloaded device; S7: According to the updated global task allocation table, migrate the migratable tasks to the task migration target device through the task migration interface; S8: Continuously monitor the load-performance ratio of edge computing devices. If the load-performance ratio exceeds the capacity threshold of the edge computing device again, re-execute S2 to S7 based on the latest load data, device performance parameters, and network bandwidth information, wherein the load data includes CPU utilization, memory occupancy, and storage space utilization, until the load-performance ratios of all edge computing devices are within the capacity threshold of the edge computing devices.

2. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: It also includes S9, which uses a deep reinforcement learning algorithm to design a reward function. The calculation formula of the reward function is: (1) in, represents the reward function, , and are the corresponding weight coefficients, Indicates the load balancing effect of the entire network. Indicates the quality of task execution, Indicates network bandwidth utilization; The optimal task allocation strategy is learned through the reward function. The learned task allocation strategy is used to dynamically adjust the task allocation according to the real-time edge computing device load performance ratio, task execution quality and network bandwidth utilization. If a new edge computing device is connected to the power distribution network or the device performance parameters change, the updated network topology data and device data are obtained, and S1 to S8 are re-executed.

3. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: In S1, the graph database Neo4j is used to build a graph model of topological structure data.

4. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: In S2, the current load data, device performance parameters, and network bandwidth information of each edge computing device are obtained through network monitoring tools.

5. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: The load performance ratio calculation formula in S2 is: (2) in, Indicates the load performance ratio, , , represents the weight coefficient, Indicates the CPU utilization, Indicates the memory usage. Indicates the storage space usage. Indicates device performance parameters.

6. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: In S4, Neo4j's graph search algorithm is used to search for neighboring devices connected to each device in the overloaded device set.

7. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: In S5, the load data of the overloaded equipment set is used to divide the migration tasks into migratable tasks and non-migratable tasks using the K-means clustering algorithm.

8. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: In S6, a greedy algorithm is used to generate a task migration plan for migratable tasks. Each time, a migratable task with the highest load-performance ratio and the smallest resource requirement is selected and assigned to the migration target device with the best load-performance ratio and large resources, and the global task allocation table is updated.

9. The edge computing load balancing method for a power distribution network according to claim 1, characterized in that: The global task allocation table updated in S7 migrates the migratable tasks to the task migration target device through the task migration interface of Kubernetes.

10. The edge computing load balancing method of a power distribution network according to claim 2, characterized in that: In S9, the Q-Learning reinforcement learning algorithm is used to optimize the load balancing effect of the entire network, the quality of task execution and the utilization of network bandwidth.

Images