A resource scheduling method and system based on power internet gateway edge computing

Abstract

The application provides a resource scheduling method and system based on power Internet of Things gateway edge computing, the method comprises the following steps: obtaining power target tasks are split into several power sub-tasks; determining the adaptation degree between each power sub-task and each task processing node, and determining the processing node set for each power sub-task according to the adaptation degree; constructing an optimization model based on the power sub-task and the processing node set, solving the target scheduling model to obtain the best processing node corresponding to each power sub-task; extracting the task information of the power sub-task and the state information of the best processing node, determining the scheduling queue based on the task information and the state information to obtain the best scheduling queue, and executing the power target task based on the best scheduling queue to complete the resource scheduling process. The application can effectively shorten the task execution time and improve the task completion efficiency, so as to realize the resource scheduling process.

Description

Technical Field

[0001] This invention belongs to the technical field of cloud, local, and edge service resource scheduling, and specifically relates to a resource scheduling method and system based on edge computing of the power Internet of Things. Background Technology

[0002] Power IoT gateways, as a special type of IoT gateway, combine the characteristics of edge computing, enabling them to perform tasks such as data processing, protocol conversion, and filtering at the network edge. This type of gateway is typically deployed close to IoT devices to reduce latency and reliance on the cloud, while still performing data processing, protocol conversion, and filtering at the edge, making it ideal for applications requiring real-time response and local decision-making.

[0003] In traditional cloud computing models, the practice of transmitting all tasks to the cloud for processing has gradually revealed its bottlenecks and drawbacks. To expand the application of cloud computing, edge computing, as an auxiliary technology to meet the needs of intelligent scenarios, has attracted much attention, offering significant advantages in reducing network congestion and transmission latency, and improving service quality. However, due to limitations imposed by environmental and energy consumption factors, the available resources are relatively limited. Therefore, how to achieve higher-quality services under limited resource conditions has become a pressing challenge.

[0004] For existing technologies, the cloud-edge-device collaborative computing architecture is characterized by a large number of devices, diverse resources, and complex hierarchical structures. Therefore, existing technologies cannot flexibly determine the offloading of computing and resource allocation, nor can they achieve intelligent scheduling. This results in problems such as long processing time, high energy consumption, and low efficiency in the actual processing of tasks. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a resource scheduling method and system based on gateway edge computing in the power Internet of Things, which solves the technical problems in the prior art.

[0006] In a first aspect, the present invention provides the following technical solution: a resource scheduling method based on gateway edge computing in the power Internet of Things, comprising:

[0007] Obtain the power target task issued by the user, and break down the power target task into several power sub-tasks;

[0008] Determine the fit between each power subtask and each task processing node, and determine a set of processing nodes for each power subtask based on the fit;

[0009] An optimization model is constructed based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model. The target scheduling model is then solved to obtain the optimal processing node for each power sub-task. The optimal processing node is a local node, a power Internet of Things (IoT) node, or a cloud node.

[0010] Extract the task information of the power subtask and the status information of the optimal processing node, determine the scheduling queue based on the task information and the status information to obtain the optimal scheduling queue, and execute the power target task based on the optimal scheduling queue to complete the resource scheduling process.

[0011] Compared with existing technologies, the beneficial effects of this invention are as follows: First, this invention obtains the power target task issued by the user and breaks it down into several power sub-tasks. Then, it determines the fit between each power sub-task and each task processing node, and determines a set of processing nodes for each power sub-task based on the fit. Next, it constructs an optimization model based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model, solves the target scheduling model to obtain the optimal processing node for each power sub-task, and finally extracts the task information of the power sub-tasks and the status information of the optimal processing node. Based on the task information and status information, it determines the scheduling queue to obtain the optimal scheduling queue, and executes the power target task based on the optimal scheduling queue to complete the resource allocation. In the source scheduling process, this invention first determines the set of processing nodes based on adaptability. By considering the relationships and importance between different resource indicators of different nodes, the corresponding set of processing nodes is determined, so that the determined set of processing nodes can handle tasks well and avoid resource waste and over-concentration. Then, a target scheduling model is constructed and the optimal processing node is determined. Taking into account the differences in latency sensitivity and energy consumption characteristics of different sub-tasks in cloud-edge-device collaboration scenarios, with the goal of minimizing latency and energy consumption, the task offloading decision, channel allocation and edge gateway computing resources are jointly optimized, which can effectively reduce latency and energy consumption. Finally, the task is executed through the optimal scheduling queue, which can effectively shorten the task execution time and improve the task completion efficiency, thus realizing the resource scheduling process.

[0012] Preferably, the step of determining the fit between each power subtask and each task processing node, and determining a set of processing nodes for each power subtask based on the fit, includes:

[0013] One of the power subtasks is pre-scheduled to the task processing node. The task processing node is then extracted. Upper Resource usage of a type of resource And based on resource usage Build a resource usage matrix ;

[0014] Resource usage matrix The elements in the data are normalized to obtain normalized resource usage. :

[0015] ;

[0016] In the formula, Resource Usage Matrix The Middle row vectors. They represent the first These resources are categorized as positive or negative indicators.

[0017] Based on the normalized resource usage Calculate resource entropy :

[0018] ;

[0019] Based on the resource entropy value Calculate resource weights :

[0020] ;

[0021] Based on the resource weight Determine the set of processing nodes.

[0022] Preferably, the method based on the resource weight The steps to determine the set of processing nodes include:

[0023] Based on the resource weight Construct the target adaptation matrix :

[0024] ;

[0025] In the formula, For the target adaptation matrix, the first Line number Column elements;

[0026] Based on the target adaptation matrix Calculate the first fitness set With the second fitness set :

[0027] ;

[0028] ;

[0029] In the formula, The first First fit, second fit, The target adaptation matrix is ​​respectively The Middle Column vectors;

[0030] Based on the first fitness set With the second fitness set Calculate the final fit :

[0031] ;

[0032] Determine the final fit degree for each power subtask, sort the task processing nodes in descending order according to the final fit degree, and select the top few task processing nodes to store in the corresponding processing node set.

[0033] Preferably, in the step of constructing an optimization model based on the power subtask and the set of processing nodes to obtain the target scheduling model, the target scheduling model is:

[0034] ;

[0035] ;

[0036] ;

[0037] , ;

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] In the formula, They are respectively The power subtask corresponding to the local node of the time period Offloading decisions, locally allocated computing resources, and selection of power IoT nodes. Channels, at key points of the power Internet of Things Allocated computing resources For the total time period, The first and second weights are respectively. The total task latency and total task energy consumption are... Representing the power subtasks respectively Executed at local nodes, power IoT nodes, and cloud nodes. They are respectively power sub-tasks Local processing latency, power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Execution time and power consumption of sub-tasks The time required for offloading from the local node to the cloud node. They are respectively power sub-tasks Local execution of energy consumption and power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Wireless transmission power consumption, power subtask Energy consumption when offloading from a local node to a cloud node. For uninstallation strategy, These are local, gateway, and cloud, respectively. For power sub-tasks Execution latency, For power sub-tasks Maximum tolerable latency, Key nodes of the power Internet of Things The channel, , These are the local maximum computing resources and key nodes in the power Internet of Things. The maximum computing resources.

[0043] Preferably, the step of solving the target scheduling model to obtain the optimal processing node corresponding to each power subtask includes:

[0044] Construct an initial solution network, taking network entities in the target scheduling model as nodes, communication relationships between network entities as edges, and the state space of power subtasks as node attributes, and establish a graph data structure;

[0045] In each iteration, several graph convolution operations are performed on the nodes in the graph data structure, and the state of each node is calculated. :

[0046] ;

[0047] In the formula, Indicates the first The state of the nodes output by the layer graph convolutional layer. It is a non-linear activation function. For the first The aggregation function of the layered graph convolutional layer. For neighboring nodes state, For the first The edge set of a layer graph convolutional layer The weight of the edge. For embedding summation operations;

[0048] Based on state Calculate network loss :

[0049] ;

[0050] In the formula, It is a multilayer perceptron;

[0051] Adjust edge weights and minimize network loss To obtain an optimized graph data structure, new node attributes are extracted from the optimized graph data structure. ;

[0052] Output the target value based on the new node attributes. And calculate the final loss. :

[0053] ;

[0054] ;

[0055] In the formula, For the first The action space of the power subtask at the time step. For network parameters, For the updated network parameters, For the reward function, For discount parameters, For the first The target value at each time step They represent the first New node attributes of the time step, action space of the power subtask;

[0056] By minimizing the final loss and iteratively updating the network parameters until the iteration stopping condition is met, the optimal graph data network structure is output. The target value of the optimal graph data network structure in the last iteration is output, and the action space of the power sub-task is extracted from the target value of the optimal graph data network structure in the last iteration to obtain the optimal processing node corresponding to each power sub-task.

[0057] Preferably, the step of extracting the task information of the power subtask and the status information of the optimal processing node, and determining the scheduling queue based on the task information and the status information to obtain the optimal scheduling queue includes:

[0058] Extract the task information of the power subtask and the status information of the optimal processing node, take the power subtask as the target node, and aggregate the task information and the status information based on the target node to obtain a summarized embedding. :

[0059] :

[0060] In the formula, For the target node Feature embedding of task information corresponding to successor nodes, These are the nonlinear transformation functions for the first eigenvector and the second eigenvector, respectively. For the target node Feature embedding of corresponding state information;

[0061] Obtain the preset scheduling network and embed the summary of each target node as the input state. Based on the preset scheduling network, a preset scheduling queue set and a reward value are determined. And calculate the action evaluation value. :

[0062] ;

[0063] ;

[0064] In the formula, , The average completion time of tasks across all scheduling queues, and the completion time of tasks in the current scheduling queue. , For the node resource utilization of all scheduling queues, and the node resource utilization of the current scheduling queue, , These are the value state network and the advantage function network in the preset scheduling network, respectively. , These are the value state network parameters and the advantage function network parameters, respectively. The scheduling queue is a pre-defined set of scheduling queues;

[0065] The scheduling queue that maximizes the action evaluation value is found in the set of scheduling queues to obtain the scheduling queue to be optimized. And based on the scheduling queue to be optimized The input state is updated to obtain the updated input state. ;

[0066] The input state Reward Value Queues to be optimized Update input status Store the data to obtain the target set. ;

[0067] Based on the target set Determine the optimal scheduling queue.

[0068] Preferably, the step of extracting the task information of the power subtask and the status information of the optimal processing node, and determining the scheduling queue based on the task information and the status information to obtain the optimal scheduling queue includes:

[0069] Based on the target set Calculate the first predicted value Compared with the second predicted value :

[0070] ;

[0071] ;

[0072] ;

[0073] In the formula, This is a set of network parameters, including value state network parameters and advantage function network parameters. Indicates that the input is Action evaluation value, The current best scheduling queue, Indicates that the input is Action evaluation value, Indicates that the input is Action evaluation value, For the new network parameter set;

[0074] Based on the first predicted value Compared with the second predicted value Calculate the predicted loss value :

[0075] ; ;

[0076] In the formula, For the first The target error value of the step. For the first The predicted loss value of the step. Discount factor;

[0077] Based on the predicted loss value For the set of network parameters in the preset scheduling network New network parameter set Perform gradient descent updates, stop iteration when the iteration stopping condition is met, and output the optimized scheduling network:

[0078] ; ;

[0079] In the formula, For learning rate, For the gradient of the parameters, This is the soft update coefficient. These are the updated set of network parameters and the new set of network parameters, respectively.

[0080] The optimal scheduling queue is output based on the optimized scheduling network.

[0081] Secondly, the present invention provides the following technical solution: a resource scheduling system based on gateway edge computing in the power Internet of Things, the system comprising:

[0082] The splitting module is used to obtain the power target task issued by the user and split the power target task into several power sub-tasks;

[0083] The adaptation module is used to determine the compatibility degree between each power subtask and each task processing node, and to determine a set of processing nodes for each power subtask based on the compatibility degree.

[0084] The construction module is used to construct an optimization model based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model, and to solve the target scheduling model to obtain the optimal processing node corresponding to each power sub-task, wherein the optimal processing node is a local node, a power Internet of Things node, or a cloud node.

[0085] The scheduling module is used to extract the task information of the power sub-task and the status information of the best processing node, determine the scheduling queue based on the task information and the status information to obtain the best scheduling queue, and execute the power target task based on the best scheduling queue to complete the resource scheduling process.

[0086] Thirdly, the present invention provides the following technical solution: a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the resource scheduling method based on the power Internet of Things edge computing as described above.

[0087] Fourthly, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the resource scheduling method based on the gateway edge computing of the power Internet of Things as described above. Attached Figure Description

[0088] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0089] Figure 1 A flowchart of a resource scheduling method based on gateway edge computing of the power Internet of Things provided in Embodiment 1 of the present invention;

[0090] Figure 2 This is a structural block diagram of a resource scheduling system based on gateway edge computing of the power Internet of Things provided in Embodiment 2 of the present invention;

[0091] Figure 3 This is a schematic diagram of the hardware structure of a computer provided for another embodiment of the present invention.

[0092] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0093] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0094] Example 1

[0095] In Embodiment 1 of the present invention, as Figure 1 As shown, a resource scheduling method based on gateway edge computing in the power Internet of Things includes:

[0096] S1. Obtain the power target task issued by the user, and break down the power target task into several power sub-tasks;

[0097] Specifically, the power target task here can be issued by the user through the user terminal, which is the local node mentioned below. The splitting process here can be achieved through the task decomposition method in project management or through the large model in existing technology.

[0098] S2. Determine the compatibility between each power subtask and each task processing node, and determine a set of processing nodes for each power subtask based on the compatibility.

[0099] Here, the task processing node is specifically an edge node, i.e., a key point in the power Internet of Things. As for the local node and the cloud node, the local node is the user's terminal device. That is, by default, the user issues tasks on the current terminal device. Therefore, there is only one local node, so there is no need to determine the adaptability. It is only necessary to choose whether to execute locally. As for the cloud node, it is considered to have unlimited storage space and can store all service models. Therefore, in this application, it is also assumed that there is only one cloud node. However, for the gateway, there can actually be several nodes. By default, a subtask can only be executed on one edge node. This restriction does not exist for local nodes and cloud nodes.

[0100] Step S2 includes:

[0101] S21. Determine one of the power subtasks to be pre-scheduled to the task processing node. The task processing node is then extracted. Upper Resource usage of a type of resource And based on resource usage Build a resource usage matrix ;

[0102] Specifically, the resources here can be CPU resources, memory resources, network resources, and disk resources.

[0103] S22, Resource Usage Matrix The elements in the data are normalized to obtain normalized resource usage. :

[0104] ;

[0105] In the formula, Resource Usage Matrix The Middle row vectors. They represent the first These resources are categorized as positive or negative indicators.

[0106] Specifically, for positive indicators, the higher the value, the better; for negative indicators, the lower the value, the better. By normalizing the values, multiple resources can be considered more accurately.

[0107] S23. Based on the normalized resource usage Calculate resource entropy :

[0108] ;

[0109] S24. Based on the resource entropy value Calculate resource weights :

[0110] ;

[0111] Specifically, resource scheduling typically involves multiple decision objectives, such as CPU, memory, network bandwidth, and disk I / O. By calculating the weight of each indicator and weighting them according to the relative importance of the decision objectives, this method ensures that important decision objectives receive higher weights in the calculation, thus more accurately reflecting their impact on the resource scheduling decision outcome. Simultaneously, by identifying and evaluating the correlations and importance among indicators, redundant and duplicate indicators can be effectively eliminated or reduced, improving the efficiency and accuracy of decision-making.

[0112] S25, Based on the resource weights Determine the set of processing nodes;

[0113] Step S25 includes:

[0114] S251, Based on the resource weights Construct the target adaptation matrix :

[0115] ;

[0116] In the formula, For the target adaptation matrix, the first Line number The elements of the column.

[0117] S252, Based on the target adaptation matrix Calculate the first fitness set With the second fitness set :

[0118] ;

[0119] ;

[0120] In the formula, The first First fit, second fit, The target adaptation matrix is ​​respectively The Middle Column vectors.

[0121] S253, Based on the first fitness set With the second fitness set Calculate the final fit :

[0122] .

[0123] S254. Determine the final fit degree corresponding to each power subtask, sort the task processing nodes in descending order according to the size of the final fit degree, and select the first few task processing nodes to store in the corresponding processing node set.

[0124] Specifically, the final fit is for a single subtask. By performing the above steps on all edge subtasks, the fit of each subtask is obtained. The higher the fit, the more suitable the node is for executing the subtask. In practice, there may be a situation where one node corresponds to multiple subtasks, which is allowed. That is, multiple subtasks can be executed on one node. However, if one subtask corresponds to multiple nodes, the nodes need to be filtered according to the priority of the subtask. That is, the same nodes corresponding to higher priority are retained, while the same nodes corresponding to lower priority are removed. Then, another node is selected from the sorted node sequence.

[0125] S3. Based on the power sub-tasks and the set of processing nodes, an optimization model is constructed to obtain a target scheduling model. The target scheduling model is solved to obtain the optimal processing node corresponding to each power sub-task. The optimal processing node is a local node, a power Internet of Things node, or a cloud node.

[0126] Specifically, the purpose of this modeling is to ensure that each electronic subtask corresponds to a node, which can be a local node, a power IoT key point, or a cloud node. This is the offloading decision process, which allows each power subtask to be executed locally, at the edge, or in the cloud.

[0127] The target scheduling model is as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] , ;

[0132] ;

[0133] ;

[0134] ;

[0135] ;

[0136] In the formula, They are respectively The power subtask corresponding to the local node of the time period Offloading decisions, locally allocated computing resources, and selection of power IoT nodes. Channels, at key points of the power Internet of Things Allocated computing resources For the total time period, The first and second weights are respectively. The total task latency and total task energy consumption are... Representing the power subtasks respectively Executed at local nodes, power IoT nodes, and cloud nodes. They are respectively power sub-tasks Local processing latency, power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Execution time and power consumption of sub-tasks The time required for offloading from the local node to the cloud node. They are respectively power sub-tasks Local execution of energy consumption and power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Wireless transmission power consumption, power subtask Energy consumption when offloading from a local node to a cloud node. For uninstallation strategy, These are local, gateway, and cloud, respectively. For power sub-tasks Execution latency, For power sub-tasks Maximum tolerable latency, Key nodes of the power Internet of Things The channel, , These are the local maximum computing resources and key nodes in the power Internet of Things. The maximum computing resources;

[0137] Specifically, this model minimizes the total latency and energy consumption of tasks by optimizing task offloading decisions, local computing resource allocation, transmission channel allocation, and edge node computing resource allocation.

[0138] Step S3 includes:

[0139] S31. Construct an initial solution network, taking the network entities in the target scheduling model as nodes, the communication relationships between network entities as edges, the state space of the power subtask as node attributes, and establishing a graph data structure.

[0140] Specifically, in this application, the initial solution network is a GNN-DQN network. For graph data results, if there is network communication between network entities, there is an edge between the two entities. Therefore, for the initial solution network, its input is the nodes, edges, and node representation matrix of the graph data structure.

[0141] S32. In each iteration, perform several graph convolution operations on the nodes in the graph data structure and calculate the state of each node. :

[0142] ;

[0143] In the formula, Indicates the first The state of the nodes output by the layer graph convolutional layer. It is a non-linear activation function. For the first The aggregation function of the layered graph convolutional layer. Neighboring nodes state, For the first The edge set of a layer graph convolutional layer The weight of the edge. For embedding summation operations;

[0144] Specifically, the non-linear activation function here is the ReLU function, and the aggregation function is used to map the information of the neighboring nodes of a node into a vector, such as mean aggregation and attention aggregation. As for the state of the node in the first layer, the node's own state can be used.

[0145] S33, State-based Calculate network loss :

[0146] ;

[0147] In the formula, It is a multilayer perceptron;

[0148] S34. Adjust the edge weights and minimize the network loss. To obtain an optimized graph data structure, new node attributes are extracted from the optimized graph data structure. ;

[0149] Specifically, the new node attributes here refer to the state of the nodes output in the optimized graph data structure.

[0150] S35. Output the target value based on the new node attributes. And calculate the final loss. :

[0151] ;

[0152] ;

[0153] In the formula, For the first The action space of the power subtask at the time step. For network parameters, For the updated network parameters, For the reward function, For discount parameters, For the first The target value at each time step They represent the first New node attributes for time steps, and the action space for power subtasks.

[0154] S36. By minimizing the final loss and iteratively updating the network parameters until the iteration stopping condition is met, the optimal graph data network structure is output, the target value of the optimal graph data network structure of the last iteration is output, and the action space of the power sub-task is extracted from the target value of the optimal graph data network structure of the last iteration, so as to obtain the optimal processing node corresponding to each power sub-task.

[0155] Specifically, the action space is actually the target to be optimized in the above model. The final action space can be extracted from the target value of the optimal graph data network structure in the last iteration. The action space contains the targets of unloading decisions and resource scheduling, which can determine the best processing node, that is, whether each subtask is executed locally, at the edge, or in the cloud.

[0156] S4. Extract the task information of the power sub-task and the status information of the best processing node, determine the scheduling queue based on the task information and the status information to obtain the best scheduling queue, and execute the power target task based on the best scheduling queue to complete the resource scheduling process.

[0157] Step S4 includes:

[0158] S41. Extract the task information of the power subtask and the status information of the optimal processing node, take the power subtask as the target node, and aggregate the task information and the status information based on the target node to obtain a summarized embedding. :

[0159] :

[0160] In the formula, For the target node Feature embedding of task information corresponding to successor nodes, These are the nonlinear transformation functions for the first eigenvector and the second eigenvector, respectively. For the target node Feature embedding of corresponding state information;

[0161] Specifically, the task information here includes the task's execution status, whether scheduling is supported, and the task's dependencies. The status information includes the node's original feature information and the node's resource usage. The first feature vector nonlinear transformation function and the second feature vector nonlinear transformation function are tanh and ReLU functions, respectively.

[0162] S42. Obtain the preset scheduling network, and embed the summary of each target node as the input state. Based on the preset scheduling network, a preset scheduling queue set and a reward value are determined. And calculate the action evaluation value. :

[0163] ;

[0164] ;

[0165] In the formula, , The average completion time of tasks across all scheduling queues, and the completion time of tasks in the current scheduling queue. , For the node resource utilization of all scheduling queues, and the node resource utilization of the current scheduling queue, , These are the value state network and the advantage function network in the preset scheduling network, respectively. , These are the value state network parameters and the advantage function network parameters, respectively. The scheduling queue is a pre-defined set of scheduling queues;

[0166] Specifically, the pre-defined scheduling network here is the DGCQN model. For this model, its output consists of two independent branches: one is responsible for estimating the scalar state value of the current system state, i.e., the value state network; the other evaluates the advantage function value of performing each action in the current state, i.e., the advantage function network parameter. The dimension of the advantage function value matches the dimension of the action space. This structure allows the network to learn the value of the state and the differential advantages between various actions, thereby optimizing the decision-making process.

[0167] S43. Find the scheduling queue that maximizes the action evaluation value in the set of scheduling queues to obtain the scheduling queue to be optimized. And based on the scheduling queue to be optimized The input state is updated to obtain the updated input state. .

[0168] S44, Set the input state Reward Value Queues to be optimized Update input status Store the data to obtain the target set. .

[0169] S45. Based on the target set Determine the optimal scheduling queue;

[0170] Step S45 includes:

[0171] S451, Based on the target set Calculate the first predicted value Compared with the second predicted value :

[0172] ;

[0173] ;

[0174] ;

[0175] In the formula, This is a set of network parameters, including value state network parameters and advantage function network parameters. Indicates that the input is Action evaluation value, The current best scheduling queue, Indicates that the input is Action evaluation value, Indicates that the input is Action evaluation value, This is the new set of network parameters.

[0176] S452, Based on the first predicted value Compared with the second predicted value Calculate the predicted loss value :

[0177] ; ;

[0178] In the formula, For the first The target error value of the step. For the first The predicted loss value of the step. This is the discount factor.

[0179] S453, Based on the predicted loss value For the set of network parameters in the preset scheduling network New network parameter set Perform gradient descent updates, stop iteration when the iteration stopping condition is met, and output the optimized scheduling network:

[0180] ; ;

[0181] In the formula, For learning rate, For the gradient of the parameters, This is the soft update coefficient. These are the updated set of network parameters and the new set of network parameters, respectively.

[0182] Specifically, for the above steps, task information and state information are abstracted into vector form. Then, through a series of information extractions, the dependency information between tasks and node state information are aggregated. This information is used as the sorting source for scheduling to generate a scheduling queue. The generated scheduling queue is executed on the optimal processing node determined in the above steps. After all tasks are completed, the total execution time of the tasks is calculated. This total execution time is used as feedback information to update the network parameters and further optimize the scheduling queue, thereby ultimately achieving the minimum total task completion time. Finally, the optimal scheduling queue is output.

[0183] S454. Output the optimal scheduling queue based on the optimized scheduling network.

[0184] The resource scheduling method based on edge computing in the power Internet of Things provided in Embodiment 1 of this invention first obtains the power target task issued by the user and breaks it down into several power sub-tasks. Then, it determines the fit between each power sub-task and each task processing node, and determines a set of processing nodes for each power sub-task based on the fit. Next, it constructs an optimization model based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model, solves the target scheduling model to obtain the optimal processing node for each power sub-task, and finally extracts the task information of the power sub-tasks and the state information of the optimal processing node. Based on the task information and state information, it determines a scheduling queue to obtain the optimal scheduling queue, and executes the power target task based on the optimal scheduling queue. To complete the resource scheduling process, this invention first determines the set of processing nodes based on adaptability. By considering the relationships and importance between different resource indicators of different nodes, the corresponding set of processing nodes is determined, ensuring that the determined set of processing nodes can handle tasks well and avoid resource waste and over-concentration. Then, a target scheduling model is constructed and the optimal processing nodes are determined. Taking into account the differences in latency sensitivity and energy consumption characteristics of different sub-tasks in cloud-edge-device collaboration scenarios, the joint optimization of task offloading decisions, channel allocation, and edge gateway computing resources is carried out with the goal of minimizing latency and energy consumption. This can effectively reduce latency and energy consumption. Finally, tasks are executed through the optimal scheduling queue, which can effectively shorten task execution time and improve task completion efficiency, thereby realizing the resource scheduling process.

[0185] Example 2

[0186] like Figure 2 As shown, in Embodiment 2 of the present invention, a resource scheduling system based on gateway edge computing of the power Internet of Things is provided. The system includes:

[0187] The splitting module 1 is used to obtain the power target task issued by the user and split the power target task into several power sub-tasks;

[0188] Adaptation module 2 is used to determine the compatibility degree between each power subtask and each task processing node, and to determine a set of processing nodes for each power subtask based on the compatibility degree.

[0189] The construction module 3 is used to construct an optimization model based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model, and to solve the target scheduling model to obtain the optimal processing node corresponding to each power sub-task, wherein the optimal processing node is a local node, a power Internet of Things node, or a cloud node.

[0190] Scheduling module 4 is used to extract the task information of the power sub-task and the status information of the best processing node, determine the scheduling queue based on the task information and the status information to obtain the best scheduling queue, and execute the power target task based on the best scheduling queue to complete the resource scheduling process.

[0191] The adapter module 2 includes:

[0192] The first matrix submodule is used to determine which of the power subtasks will be pre-scheduled to the task processing node. The task processing node is then extracted. Upper Resource usage of a type of resource And based on resource usage Build a resource usage matrix ;

[0193] The normalization submodule is used to normalize the resource usage matrix. The elements in the data are normalized to obtain normalized resource usage. :

[0194] ;

[0195] In the formula, Resource Usage Matrix The Middle row vectors. They represent the first These resources are categorized as positive or negative indicators.

[0196] The entropy submodule is used to base the normalized resource usage on... Calculate resource entropy :

[0197] ;

[0198] The weighting submodule is used to weight the resource entropy value. Calculate resource weights :

[0199] ;

[0200] The processing node submodule is used to process the resource weights. Determine the set of processing nodes.

[0201] The processing node submodule includes:

[0202] Matrix unit, used based on the resource weight Construct the target adaptation matrix :

[0203] ;

[0204] In the formula, For the target adaptation matrix, the first Line number Column elements;

[0205] Adaptation unit, used to adapt to the target adaptation matrix Calculate the first fitness set With the second fitness set :

[0206] ;

[0207] ;

[0208] In the formula, The first First fit, second fit, The target adaptation matrix is ​​respectively The Middle Column vectors;

[0209] The final fitness unit is used to determine the fitness level based on the first fitness set. With the second fitness set Calculate the final fit :

[0210] ;

[0211] The sorting unit is used to determine the final fitness degree corresponding to each power subtask, sort the task processing nodes in descending order according to the size of the final fitness degree, and select the top several task processing nodes to store in the corresponding processing node set.

[0212] The construction module 3 includes:

[0213] A submodule is constructed to build the initial solution network, taking the network entities in the target scheduling model as nodes, the communication relationships between network entities as edges, the state space of the power subtask as node attributes, and establishing a graph data structure.

[0214] The convolution submodule is used to perform several graph convolution operations on the nodes in the graph data structure and calculate the state of each node in each iteration. :

[0215] ;

[0216] In the formula, Indicates the first The state of the nodes output by the layer graph convolutional layer. It is a non-linear activation function. For the first The aggregation function of the layered graph convolutional layer. Neighboring nodes state, For the first The edge set of a layer graph convolutional layer The weight of the edge. For embedding summation operations;

[0217] The network loss submodule is used for state-based... Calculate network loss :

[0218] ;

[0219] In the formula, It is a multilayer perceptron;

[0220] The attribute submodule is used to adjust edge weights and minimize network loss. To obtain an optimized graph data structure, new node attributes are extracted from the optimized graph data structure. ;

[0221] The final submodule is used to output the target value based on the new node attributes. And calculate the final loss. :

[0222] ;

[0223] ;

[0224] In the formula, For the first The action space of the power subtask at the time step. For network parameters, For the updated network parameters, For the reward function, For discount parameters, For the first The target value at each time step They represent the first New node attributes of the time step, action space of the power subtask;

[0225] The extraction submodule is used to output the optimal graph data network structure by minimizing the final loss and iteratively updating the network parameters until the iteration stopping condition is met. It outputs the target value of the optimal graph data network structure in the last iteration and extracts the action space of the power sub-task from the target value of the optimal graph data network structure in the last iteration to obtain the optimal processing node corresponding to each power sub-task.

[0226] The scheduling module 4 includes:

[0227] The aggregation submodule is used to extract the task information of the power subtask and the status information of the optimal processing node, take the power subtask as the target node, and aggregate the task information and the status information based on the target node to obtain a summarized embedding. :

[0228] :

[0229] In the formula, For the target node Feature embedding of task information corresponding to successor nodes, These are the nonlinear transformation functions for the first eigenvector and the second eigenvector, respectively. For the target node Feature embedding of corresponding state information;

[0230] The evaluation submodule is used to obtain the preset scheduling network and embed the summary of each target node as the input state. Based on the preset scheduling network, a preset scheduling queue set and a reward value are determined. And calculate the action evaluation value. :

[0231] ;

[0232] ;

[0233] In the formula, , The average completion time of tasks across all scheduling queues, and the completion time of tasks in the current scheduling queue. , For the node resource utilization of all scheduling queues, and the node resource utilization of the current scheduling queue, , These are the value state network and the advantage function network in the preset scheduling network, respectively. , These are the value state network parameters and the advantage function network parameters, respectively. The scheduling queue is a pre-defined set of scheduling queues;

[0234] The update submodule is used to find the scheduling queue that maximizes the action evaluation value in the set of scheduling queues, so as to obtain the scheduling queue to be optimized. And based on the scheduling queue to be optimized The input state is updated to obtain the updated input state. ;

[0235] The collection submodule is used to store the input states. Reward Value Queues to be optimized Update input status Store the data to obtain the target set. ;

[0236] The queue scheduling submodule is used to schedule based on the target set. Determine the optimal scheduling queue.

[0237] The queue scheduling submodule includes:

[0238] Prediction unit, used for prediction based on the target set Calculate the first predicted value Compared with the second predicted value :

[0239] ;

[0240] ;

[0241] ;

[0242] In the formula, This is a set of network parameters, including value state network parameters and advantage function network parameters. Indicates that the input is Action evaluation value, The current best scheduling queue, Indicates that the input is Action evaluation value, Indicates that the input is Action evaluation value, For the new network parameter set;

[0243] Prediction loss unit, used to predict loss based on the first predicted value Compared with the second predicted value Calculate the predicted loss value :

[0244] ; ;

[0245] In the formula, For the first The target error value of the step. For the first The predicted loss value of the step. Discount factor;

[0246] An optimization unit is configured to optimize based on the predicted loss value. For the set of network parameters in the preset scheduling network New network parameter set Perform gradient descent updates, stop iteration when the iteration stopping condition is met, and output the optimized scheduling network:

[0247] ; ;

[0248] In the formula, For learning rate, For the gradient of the parameters, This is the soft update coefficient. These are the updated set of network parameters and the new set of network parameters, respectively.

[0249] The queue output unit is used to output the optimal scheduling queue based on the optimized scheduling network.

[0250] In other embodiments of the present invention, the present invention provides the following technical solution: a computer, including a memory 102, a processor 101, and a computer program stored in the memory 102 and executable on the processor 101, wherein the processor 101 executes the computer program to implement the resource scheduling method based on the gateway edge computing of the power Internet of Things as described above.

[0251] Specifically, the processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0252] The memory 102 may include a large-capacity memory for data or instructions. For example, and not limitingly, the memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 102 may include removable or non-removable (or fixed) media. Where appropriate, the memory 102 may be internal or external to a data processing device. In a particular embodiment, the memory 102 is non-volatile memory. In a particular embodiment, the memory 102 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0253] The memory 102 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 101.

[0254] The processor 101 reads and executes the computer program instructions stored in the memory 102 to implement the above-mentioned resource scheduling method based on the gateway edge computing of the power Internet of Things.

[0255] In some embodiments, the computer may further include a communication interface 103 and a bus 100. For example, Figure 3 As shown, the processor 101, memory 102, and communication interface 103 are connected through bus 100 and complete communication with each other.

[0256] The communication interface 103 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 103 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0257] Bus 100 includes hardware, software, or both, that couples components of a computer device together. Bus 100 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 100 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 100 may include one or more buses. Although specific buses are described and illustrated in the embodiments of the present invention, the present invention is contemplated by any suitable bus or interconnect.

[0258] The computer can execute the resource scheduling method based on the power Internet of Things gateway edge computing of the present invention based on the resource scheduling system based on the power Internet of Things gateway edge computing, thereby realizing resource scheduling based on the power Internet of Things gateway edge computing.

[0259] In some further embodiments of the present invention, in conjunction with the above-described resource scheduling method based on gateway edge computing of the power Internet of Things, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described resource scheduling method based on gateway edge computing of the power Internet of Things.

[0260] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0261] More specific examples of readable media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0262] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0263] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0264] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A resource scheduling method based on gateway edge computing in the power Internet of Things, characterized in that, include: Obtain the power target task issued by the user, and break down the power target task into several power sub-tasks; Determine the fit between each power subtask and each task processing node, and determine a set of processing nodes for each power subtask based on the fit; An optimization model is constructed based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model. The target scheduling model is then solved to obtain the optimal processing node for each power sub-task. The optimal processing node is a local node, a power Internet of Things (IoT) node, or a cloud node. Extract the task information of the power subtask and the status information of the best processing node, determine the scheduling queue based on the task information and the status information to obtain the best scheduling queue, and execute the power target task based on the best scheduling queue to complete the resource scheduling process; The step of extracting the task information of the power subtask and the status information of the optimal processing node, and determining the scheduling queue based on the task information and the status information to obtain the optimal scheduling queue includes: Extract the task information of the power subtask and the status information of the optimal processing node, take the power subtask as the target node, and aggregate the task information and the status information based on the target node to obtain a summarized embedding. : ： In the formula, For the target node Feature embedding of task information corresponding to successor nodes, These are the nonlinear transformation functions for the first eigenvector and the second eigenvector, respectively. For the target node Feature embedding of corresponding state information; Obtain the preset scheduling network and embed the summary of each target node as the input state. Based on the preset scheduling network, a preset scheduling queue set and a reward value are determined. And calculate the action evaluation value. : ； ； In the formula, , The average completion time of tasks across all scheduling queues, and the completion time of tasks in the current scheduling queue. , For the node resource utilization of all scheduling queues, and the node resource utilization of the current scheduling queue, , These are the value state network and the advantage function network in the preset scheduling network, respectively. , These are the value state network parameters and the advantage function network parameters, respectively. The scheduling queue is a pre-defined set of scheduling queues; Find the action evaluation value in the set of scheduling queues. The largest scheduling queue is used to obtain the scheduling queue to be optimized. And based on the scheduling queue to be optimized The input state is updated to obtain the updated input state. ; The input state Reward Value Queues to be optimized Update input status Store the data to obtain the target set. ; Based on the target set Determine the optimal scheduling queue.

2. The resource scheduling method based on gateway edge computing of the power Internet of Things according to claim 1, characterized in that, The step of determining the fit between each power subtask and each task processing node, and determining a set of processing nodes for each power subtask based on the fit, includes: One of the power subtasks is pre-scheduled to the task processing node. The task processing node is then extracted. Upper Resource usage of a type of resource And based on resource usage Build a resource usage matrix ; Resource usage matrix The elements in the data are normalized to obtain normalized resource usage. : ； In the formula, Resource Usage Matrix The Middle row vectors. They represent the first These resources are categorized as positive or negative indicators. Based on the normalized resource usage Calculate resource entropy : ； Based on the resource entropy value Calculate resource weights : ； Based on the resource weight Determine the set of processing nodes.

3. The resource scheduling method based on gateway edge computing of the power Internet of Things according to claim 2, characterized in that, Based on the resource weight The steps to determine the set of processing nodes include: Based on the resource weight Construct the target adaptation matrix : ； In the formula, For the target adaptation matrix, the first Line number Column elements; Based on the target adaptation matrix Calculate the first fitness set With the second fitness set : ； ； In the formula, The first First fit, second fit, The target adaptation matrix is ​​respectively The Middle Column vectors; Based on the first fitness set With the second fitness set Calculate the final fit : ； Determine the final fit degree for each power subtask, sort the task processing nodes in descending order according to the final fit degree, and select the top few task processing nodes to store in the corresponding processing node set.

4. The resource scheduling method based on gateway edge computing of the power Internet of Things according to claim 1, characterized in that, In the step of constructing an optimization model based on the power subtask and the set of processing nodes to obtain the target scheduling model, the target scheduling model is: ； ； ； ， ； ； ； ； ； In the formula, They are respectively The power subtask corresponding to the local node of the time period Offloading decisions, locally allocated computing resources, and selection of power IoT nodes. Channels, at key points of the power Internet of Things Allocated computing resources For the total time period, The first and second weights are respectively. The total task latency and total task energy consumption are... Representing the power subtasks respectively Executed at local nodes, power IoT nodes, and cloud nodes. They are respectively power sub-tasks Local processing latency, power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Execution time and power consumption of sub-tasks The time required for offloading from the local node to the cloud node. They are respectively power sub-tasks Local execution of energy consumption and power subtasks Offloaded from local nodes to key nodes in the power Internet of Things Wireless transmission power consumption, power subtask Energy consumption when offloading from a local node to a cloud node. For uninstallation strategy, These are local, gateway, and cloud, respectively. For power sub-tasks Execution latency, For power sub-tasks Maximum tolerable latency, Key nodes of the power Internet of Things The channel, , These are the local maximum computing resources and key nodes in the power Internet of Things. The maximum computing resources.

5. The resource scheduling method based on gateway edge computing of the power Internet of Things according to claim 1, characterized in that, The step of solving the target scheduling model to obtain the optimal processing node for each power subtask includes: Construct an initial solution network, taking network entities in the target scheduling model as nodes, communication relationships between network entities as edges, and the state space of power subtasks as node attributes, and establish a graph data structure; In each iteration, several graph convolution operations are performed on the nodes in the graph data structure, and the state of each node is calculated. : ； In the formula, Indicates the first The state of the nodes output by the layer graph convolutional layer. It is a non-linear activation function. For the first The aggregation function of the layered graph convolutional layer. For neighboring nodes state, For the first The edge set of a layer graph convolutional layer The weight of the edge. For embedding summation operations; Based on state Calculate network loss : ； In the formula, It is a multilayer perceptron; Adjust edge weights and minimize network loss To obtain an optimized graph data structure, new node attributes are extracted from the optimized graph data structure. ; Output the target value based on the new node attributes. And calculate the final loss. : ； ； In the formula, For the first The action space of the power subtask at the time step. For network parameters, For the updated network parameters, For the reward function, For discount parameters, For the first The target value at each time step They represent the first New node attributes of the time step, action space of the power subtask; By minimizing the final loss and iteratively updating the network parameters until the iteration stopping condition is met, the optimal graph data network structure is output. The target value of the optimal graph data network structure in the last iteration is output, and the action space of the power sub-task is extracted from the target value of the optimal graph data network structure in the last iteration to obtain the optimal processing node corresponding to each power sub-task.

6. The resource scheduling method based on gateway edge computing of the power Internet of Things according to claim 1, characterized in that, The step of extracting the task information of the power subtask and the status information of the optimal processing node, and determining the scheduling queue based on the task information and the status information to obtain the optimal scheduling queue includes: Based on the target set Calculate the first predicted value Compared with the second predicted value : ； ； ； In the formula, This is a set of network parameters, including value state network parameters and advantage function network parameters. Indicates that the input is Action evaluation value, The current best scheduling queue, Indicates that the input is Action evaluation value, Indicates that the input is Action evaluation value, For the new network parameter set; Based on the first predicted value Compared with the second predicted value Calculate the predicted loss value : ； ； In the formula, For the first The target error value of the step. For the first The predicted loss value of the step. Discount factor; Based on the predicted loss value For the set of network parameters in the preset scheduling network New network parameter set Perform gradient descent updates, stop iteration when the iteration stopping condition is met, and output the optimized scheduling network: ； ； In the formula, For learning rate, For the gradient of the parameters, This is the soft update coefficient. These are the updated set of network parameters and the new set of network parameters, respectively. The optimal scheduling queue is output based on the optimized scheduling network.

7. A resource scheduling system based on gateway edge computing of the power Internet of Things, wherein the system adopts the resource scheduling method based on gateway edge computing of the power Internet of Things as described in claim 1, characterized in that, The system includes: The splitting module is used to obtain the power target task issued by the user and split the power target task into several power sub-tasks; The adaptation module is used to determine the compatibility degree between each power subtask and each task processing node, and to determine a set of processing nodes for each power subtask based on the compatibility degree. The construction module is used to construct an optimization model based on the power sub-tasks and the set of processing nodes to obtain a target scheduling model, and to solve the target scheduling model to obtain the optimal processing node corresponding to each power sub-task, wherein the optimal processing node is a local node, a power Internet of Things node, or a cloud node. The scheduling module is used to extract the task information of the power sub-task and the status information of the best processing node, determine the scheduling queue based on the task information and the status information to obtain the best scheduling queue, and execute the power target task based on the best scheduling queue to complete the resource scheduling process.

8. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the resource scheduling method based on edge computing of the power Internet of Things as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the resource scheduling method based on edge computing of the power Internet of Things as described in any one of claims 1 to 6.

Images