Heterogeneous computing power scheduling optimization method and device for intelligent operation of end-side power grid

Abstract

The invention discloses an end-edge power grid intelligent operation heterogeneous computing power scheduling optimization method and device, and belongs to the technical field of power grid intelligent operation, a power grid intelligent operation computing power scheduling dynamic game model is constructed, an edge server determines a basic pricing function according to power grid intelligent operation related information reported by power terminal equipment, and the power grid intelligent operation heterogeneous computing power scheduling dynamic game model is established; the power terminal equipment determines the data volume unloaded to the edge server and the edge server resource volume needing to be occupied according to the basic pricing function in combination with the execution demand and the cost consideration of the power terminal equipment; according to the method, the optimal unloading strategy is obtained on the basis of the dynamic game process of the power grid intelligent operation computing power scheduling dynamic game model, and reasonable distribution and scheduling of end-side computing power resources are achieved. The computing power distribution efficiency in a scene in which multiple subjects share computing power resources is improved.

Description

Technical Field

[0001] The present invention relates to a method and device for optimizing heterogeneous computing power scheduling in intelligent edge-grid operations, belonging to the technical field of intelligent grid operations. Background Art

[0002] With the development of grid intelligence, a large number of intelligent operations such as image recognition in grid inspections, data analysis in fault diagnosis, and model calculations in load forecasting are carried out in the grid system. These intelligent operations often require varying degrees of computing resource support.

[0003] However, current computing power scheduling faces many problems. On the one hand, the computing tasks of grid intelligent operations are diverse, including different types such as compute-intensive and data-intensive. Different types of tasks have very different requirements for computing power. For example, some tasks require a large amount of parallel computing power, while some have high requirements for data transmission bandwidth and storage speed. On the other hand, computing power resources are heterogeneous, with both the limited computing power of power terminal devices such as intelligent sensors and inspection robots, and the relatively powerful computing power of edge servers.

[0004] Traditional static computing power scheduling methods cannot flexibly allocate according to the dynamic changes of tasks and the real-time status of computing power resources, easily leading to resource waste or task execution delays. For example, during peak periods of grid inspections, a large number of image recognition tasks may be generated simultaneously. If the computing power of edge servers and power terminal devices cannot be reasonably scheduled, there may be a backlog of inspection image data and untimely processing, which may affect the timely discovery and handling of grid faults and seriously threaten the safe and stable operation of the grid in severe cases. At the same time, most existing scheduling methods do not consider the interest game relationship between different operation entities. In the scenario of multi-entity sharing computing power resources, resource competition conflicts may be triggered, further reducing the overall efficiency of grid intelligent operations. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for optimizing heterogeneous computing power scheduling in intelligent edge-grid operations, which rationally allocate the resource usage amounts of all participants in the grid intelligent operation scenario through dynamic game technology to improve the allocation efficiency of computing power in the scenario of multi-entity sharing computing power resources.

[0006] To achieve the above object, the present invention is implemented by the following technical solutions: In the first aspect, the present invention provides a method for optimizing heterogeneous computing power scheduling in intelligent edge-grid operations, including: Setting the edge server as the leader role in the game and the power terminal device as the follower role, and constructing a dynamic game model for computing power scheduling in grid intelligent operations, where: The edge server determines a basic pricing function based on the power grid intelligent operation related information reported by the power terminal device. The power terminal device determines the amount of data to be offloaded to the edge server and the amount of edge server resources required according to the basic pricing function, in combination with its own execution requirements and cost considerations. Based on the dynamic game process of the power grid intelligent operation computing power scheduling dynamic game model, an optimal offloading strategy is obtained to realize the reasonable allocation and scheduling of the terminal-edge computing power resources.

[0007] Further, the calculation formula of the basic pricing function is as follows: ; where, represents the basic pricing function, represents the amount of resources occupied by the power grid intelligent operation, represents the remaining resources of the edge server. When , the power grid intelligent operation can be offloaded. and are weight factors, which are positive values and are determined by the edge server.

[0008] Further, the power grid intelligent operation related information includes the local computing power resources, the power grid intelligent operation data volume, and the execution delay of the power grid intelligent operation.

[0009] Further, the calculation formula of the execution delay of the power grid intelligent operation is as follows: ; where, represents the execution delay of the power grid intelligent operation, represents the upload rate, represents the download rate, represents the data size ratio before and after processing, represents the amount of data to be calculated on the edge server, represents the power grid intelligent operation data volume, represents the amount of resources consumed by the power grid intelligent operation when processing each bit of data, represents the amount of resources occupied by the power grid intelligent operation, represents the local computing power resources.

[0010] Further, the obtaining of the optimal offloading strategy based on the dynamic game process of the power grid intelligent operation computing power scheduling dynamic game model includes: Calculating the amount of data transmitted to the edge server as the balanced data when the time for processing the power grid intelligent operation locally at the power terminal device is equal to all the time for offloading the power grid intelligent operation to the edge server for processing. , the formula is as follows: ; Calculate the minimum value of the cost function of the power terminal device according to the balance data, and the calculation formula is as follows: ; Among them, represents the offloading weight, represents the remaining resource amount of the edge server; The offloading optimization problem of the power terminal device is expressed as follows: ; For take the second derivative to find the solution that makes the smallest, and use it as the resource amount of the edge server occupied to reach the Nash equilibrium , and the calculation formula is as follows: ; The optimal offloading strategy is obtained and expressed as follows: When , ; ; ; ; When , ; Among them represents the optimal offloading resource amount, represents the optimal offloading data amount, represents the lower limit of the resource amount occupied by the edge server, represents the upper limit of the resource amount occupied by the edge server, represents the delay constraint of the grid intelligent operation.

[0011] Furthermore, the method further includes: using a supervised learning method to determine the weight factors and in the basic pricing function. Specifically: use a feedforward neural network to construct a network structure with multiple hidden layers, and train by collecting a large amount of historical data related to grid intelligent operations, so that the edge server can determine the weight factors and values according to the trained model.

[0012] In the second aspect, the present invention provides a heterogeneous computing power scheduling and optimization device for grid intelligent operations between the terminal and the edge, including: A model construction module is used to set the edge server as the leader role in the game and the power terminal device as the follower role, and construct a dynamic game model for grid intelligent operation computing power scheduling, where: The edge server determines a basic pricing function according to the grid intelligent operation related information reported by the power terminal device. The power terminal device determines the amount of data to be offloaded to the edge server and the amount of edge server resources required according to the basic pricing function, combined with its own execution requirements and cost considerations. A scheduling module is used to obtain an optimal offloading strategy based on the dynamic game process of the grid intelligent operation computing power scheduling dynamic game model, and realize the reasonable allocation and scheduling of edge-side computing power resources.

[0013] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in any one of the foregoing are realized.

[0014] In a fourth aspect, the present invention provides a computer device, including: A memory for storing computer programs / instructions; A processor for executing the computer programs / instructions to realize the steps of the method described in any one of the foregoing.

[0015] In a fifth aspect, the present invention provides a computer program product, including computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the method described in any one of the foregoing are realized.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are: The present invention provides a method and device for optimizing heterogeneous computing power scheduling of grid intelligent operations at the edge and end. By constructing a dynamic game model for grid intelligent operation computing power scheduling and obtaining an optimal offloading strategy based on the dynamic game process of the dynamic game model, the reasonable allocation and scheduling of edge-side computing power resources are realized, the overall intelligent operation efficiency is improved, the situation of overload or resource idle waste of the edge server is avoided, the resource utilization efficiency is improved, and the grid intelligent operation can operate stably and efficiently under different load conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the model architecture of the grid intelligent operation computing power scheduling system provided by an embodiment of the present invention; Figure 2 It is an overall execution flowchart of a method for optimizing heterogeneous computing power scheduling of grid intelligent operations at the edge and end provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0019] Embodiment 1. This embodiment introduces an optimization method for heterogeneous computing power scheduling in intelligent operation of the edge-side power grid, including: Set the edge server as the leader role in the game, and the power terminal device as the follower role, and construct a dynamic game model for computing power scheduling in intelligent operation of the power grid, where: The edge server determines the basic pricing function according to the information related to intelligent operation of the power grid reported by the power terminal device. The power terminal device determines the amount of data to be offloaded to the edge server and the amount of edge server resources required according to the basic pricing function, combined with its own execution requirements and cost considerations. Based on the dynamic game process of the dynamic game model for computing power scheduling in intelligent operation of the power grid, obtain the optimal offloading strategy to achieve reasonable allocation and scheduling of edge-side computing power resources.

[0020] The optimization method for heterogeneous computing power scheduling in intelligent operation of the edge-side power grid provided by this embodiment specifically involves the following steps in its application process: 1. Design of a dynamic game model for computing power scheduling in intelligent operation of the power grid; In this embodiment, the computing power scheduling in intelligent operation of the power grid is converted into a problem of offloading in intelligent operation of the power grid, and the intelligent operation of the power grid that cannot be completed locally in time is offloaded to an edge server with heterogeneous computing power.

[0021] 1. Construct a system model for computing power scheduling in intelligent operation of the power grid; The architecture of the system model for computing power scheduling in intelligent operation of the power grid is as Figure 1 shown. The present invention focuses on a scenario where multiple agents share heterogeneous computing power resources. An edge server is processing intelligent operations of the power grid offloaded by multiple power terminal devices, and there are other power terminal devices waiting to offload intelligent operations of the power grid. These power terminal devices hope to shorten the execution time of intelligent operations of the power grid by offloading them to the edge server, thereby improving the efficiency and timeliness of intelligent operations of the power grid. The main task of the edge server is to avoid its own high-load operation state while fully meeting the operation requirements of the power terminal device, and achieve precise management and efficient utilization of various heterogeneous edge server computing power resources. Intelligent operations of the power grid have the characteristic of partial offloadability, that is, some data in the operation can be offloaded to the edge server for computing and processing.

[0022] The overall execution process between the edge server and the power terminal device is as Figure 2As shown in the figure. At the initial stage of grid intelligent operation scheduling, power terminal devices first report relevant detailed information of grid intelligent operations to the edge server. This embodiment performs grid intelligent operation offloading based on the Stackelberg game principle. The edge server determines the basic pricing function according to the information about grid intelligent operations reported by the power terminal devices, and the power terminal devices determine the amount of offloaded data and the amount of resources (the amount of resources can be CPU capacity, GPU capacity, etc.) required to occupy the edge server based on this basic pricing function. When the power terminal device has a strong willingness to offload operations and the determined amount of offloaded data and occupied resources can meet the operation execution requirements, the grid intelligent operation will be transmitted to the designated edge server for processing. The cost paid by the power terminal device is jointly determined by the basic pricing function and the amount of resources occupied by the edge server.

[0023] 2. Grid intelligent operation payment model; This embodiment proposes a variable pricing scheme to solve the problems existing in the traditional grid intelligent operation computing power scheduling scheme. The basic pricing function of this scheme is a function that depends on the remaining resources of the edge server and the resources occupied by the grid intelligent operation. When a new grid intelligent operation is offloaded, the remaining resources of the edge server are known. Its definition is as follows: ; where represents the resources occupied by the grid intelligent operation, represents the remaining resources of the edge server. When , the grid intelligent operation can be offloaded. and are weight factors, which are positive values and are determined by the edge server.

[0024] The unit resource price is related to the remaining resources of the edge server and the resources occupied by the grid intelligent operation. It is a unary quadratic equation about .

[0025] ; The grid intelligent operations involved in this embodiment have the characteristic of partial offloading, allowing a certain amount of data to be transmitted to the edge server for processing. The total amount of data to be calculated for the grid intelligent operation is denoted as (in bits), and the amount of data to be calculated transmitted to the edge server is (in bits). It should be noted that different grid intelligent applications consume different amounts of resources when processing each bit of data. This amount of resources is represented by . Based on the above conditions, the processing time of the grid intelligent operation on the edge server can be expressed as: ; The payment cost of grid intelligent operation is the product of the unit resource price and the processing time of grid intelligent operation on the edge server. Then, the payment cost of grid intelligent operation can be expressed as follows: ; In this payment model, it can be observed that as the amount of offloaded data increases and the resource occupancy rate of grid intelligent operation rises, its payment cost increases accordingly, and this cost is closely related to the remaining resource amount of the edge server. When the amount of offloaded data of grid intelligent operation remains constant, if the current load of the edge server is relatively light, the growth trend of its payment cost will be relatively gentle when the resource amount occupied by grid intelligent operation on the edge server increases. On the contrary, if the load of the edge server is already at a relatively high level, any change in the resource occupancy of grid intelligent operation on the edge server may lead to significant fluctuations in its payment cost.

[0026] 3. Dynamic game modeling of each participant in computing power scheduling; In this embodiment, the edge server and the power terminal device are constructed into a Stackelberg game model. The core goal of this model is to reasonably plan the resource allocation of the edge server to ensure that the execution delay of grid intelligent operation is maintained at a relatively low level. In this model, the edge server acts as the leader, while the power terminal device acts as the follower. As the dominant force in the market, the edge server is responsible for setting the basic pricing function and optimizing the basic pricing function by changing the weight factors , The fundamental goal of the edge server is to maximize the utility function. At the same time, the power terminal device strives to minimize the cost in terms of the amount of offloaded data and resource occupancy.

[0027] When a new grid intelligent operation arrives, the power terminal device needs to report its local resource amount , the total amount of data to be calculated for the grid intelligent operation , the payment weight of the power terminal device , the delay constraint of the grid intelligent operation , the resource amount required to process 1 bit of data . The edge server counts its remaining resource amount and determines the weight factors , And the basic pricing function. In this process, the power terminal device can determine the amount of data unloaded to the edge server and the amount of resources occupied based on the payment cost. Generally, a lower payment cost means less resource occupancy of the edge server and a lower data unloading volume, but the execution delay of the grid intelligent operation may increase. On the contrary, if the power terminal device pursues a lower execution delay, it may face a higher payment cost because this usually involves more extensive resource use. Therefore, when making a choice, the power terminal device should balance the execution delay and the payment cost to find an appropriate balance point.

[0028] The delay in the execution of the grid intelligent operation refers to the part that takes longer than the local processing time of the power terminal device on the edge server.

[0029] ; where represents the execution delay of the grid intelligent operation, includes all the time from the start of transmission of the grid intelligent operation to the edge server transmitting the calculation result to the power terminal device, represents the time for the grid intelligent operation to be processed locally by the power terminal device, represents the time for data to be transmitted to the edge server, represents the time for the power terminal device to download the calculation result from the edge server.

[0030] The upload and download times of the grid intelligent operation are affected by many factors. This embodiment only focuses on the transmission rate and assumes that all conditions reach an ideal state during the upload and download processes of the grid intelligent operation. and can be expressed as follows: ; where represents the upload rate, represents the download rate, represents the ratio of the data size before and after processing.

[0031] After the grid intelligent operation is unloaded, the power terminal device will process the remaining amount of data locally, that is, . The processing time of the grid intelligent operation locally by the power terminal device can be expressed as: ; According to formulas (3), (5), (6), and (7), the execution delay of the grid intelligent operation can be further expressed as follows: ; The cost function of the power terminal device can be represented by the weighted sum of the execution delay of grid intelligent operations and the payment cost of the power terminal device: ; The payment weight of the power terminal device is always positive. This parameter reflects the degree of attention of the device to the execution delay of grid intelligent operations. If the payment weight is relatively low, it indicates that the power terminal device pays more attention to the immediacy of the current grid intelligent operations and is willing to bear additional costs for this. Conversely, when has a higher value, it means that the power terminal device can tolerate the delay of grid intelligent operations and shows a relatively passive offloading attitude.

[0032] After the power terminal device offloads the data that needs to be calculated for grid intelligent operations, the edge server will obtain corresponding benefits, and its value is equal to the payment cost of the power terminal device. What the edge server pursues is the maximization of its utility function. Once the offloaded grid intelligent operations are completed, the edge server immediately earns income. Assuming that in the ideal state, the amount of data offloaded by the power terminal device to the edge server is and the resources occupied by the grid intelligent operations on the edge server are , then the utility function of the edge server can be expressed as follows: ; The edge server plays a leading role in resource management, and its goal is to maximize the utility function. Correspondingly, the power terminal device plays the role of a follower, and its main task is to minimize the cost function. Based on this role assignment, this embodiment constructs a Stackelberg dynamic game offloading strategy to achieve reasonable scheduling of heterogeneous computing power. The core goal of this strategy is to achieve precise scheduling of edge server resources while satisfying the delay constraints of grid intelligent operations and preventing the server from bearing excessive load pressure.

[0033] II. Derivation of Nash equilibrium for grid intelligent operation computing power scheduling; This embodiment analyzes the heterogeneous computing power scheduling strategy based on the Stackelberg game model and solves the Nash equilibrium point.

[0034] 1. Balanced data; First, define that when the time for the power terminal device to locally process grid intelligent operations is equal to all the time for the grid intelligent operations to be offloaded to the edge server for processing, the amount of data transmitted to the edge server is the balanced data .

[0035] ; ; At this time, , and should be less than or equal to the latency constraint of the task .

[0036] ; That is, ; 2. Nash equilibrium point; The optimal offloading strategy for power terminal devices is to offload an appropriate amount of data and occupy an appropriate amount of edge server resources to minimize the cost function . This embodiment formulates the offloading optimization problem of power terminal devices as follows.

[0037] ; In formula (15), first, to ensure that grid intelligent operations can be offloaded, the execution latency of grid intelligent operations should be less than or equal to the latency constraint of grid intelligent operations. Offloading that does not meet this condition is useless offloading and will affect the quality of service. Second, the amount of edge server resources occupied should be less than the remaining resources of the edge server , which is determined by the objective environment. Finally, the weight factors and , the ratio of the sizes of the data before and after processing , and the offloading weight are all numbers greater than 0, which are determined by the constructed system model.

[0038] According to the balanced data, the execution latency of grid intelligent operations can be expressed as follows: ; The partial derivative of the cost function of the power terminal device with respect to the amount of data offloaded to the edge server can be expressed as follows: ; Let denote the partial derivative of with respect to the amount of data offloaded to the edge server. The shape of the cost function of the power terminal device will change according to ; If , for and , , that is, the cost function of the power terminal device is a monotonically increasing function of the amount of data unloaded to the edge server . No matter how much data the power terminal device unloads, it is difficult for the power terminal device to benefit.

[0039] If , for and , , that is, the cost function of the power terminal device is a monotonically decreasing function of the amount of data unloaded to the edge server . For and , , that is, the cost function of the power terminal device is a monotonically increasing function of the amount of data unloaded to the edge server . When , the cost function of the power terminal device ; If , for and , , that is, the cost function of the power terminal device is a constant. In this case, even if the power terminal device does not need to unload a lot of data, the power terminal device will choose to unload balanced data . For and , , that is, the cost function of the power terminal device is a monotonically increasing function of the amount of data unloaded to the edge server .

[0040] When , the cost function of the power terminal device has a minimum value, and the power terminal device can benefit from the unloading process, that is, unload balanced data. When , for , the power terminal device can also choose to unload balanced data, and when , the cost function of the power terminal device is a monotonically increasing function of the amount of data unloaded to the edge server , and the power terminal device cannot benefit from the unloading. Therefore, only when , is it possible for the power terminal device to unload. According to According to the definition, the amount of resources occupied by the power terminal device The upper limit is expressed as follows: ; Therefore, the optimal strategy of the power terminal device can be expressed by the following formula: ; If the power terminal device unloads, the amount of data unloaded is the balanced data, and the cost function of the power terminal device is the value of formula (19). At this time, the processing time of the grid intelligent operation on the local side of the power terminal device and are equal. The offloading delay of the grid intelligent operation should be less than or equal to the delay constraint of the grid intelligent operation . Combining formulas (14) and (20), the value range of the amount of resources occupied by the power terminal device can be obtained as follows: ; According to formulas (15), (19), (21), and (22), the offloading optimization problem of the power terminal device is further expressed as follows: ; In formula (23), the amount of data unloaded by the power terminal device to the edge server has been determined , but the amount of resources that the power terminal device needs to occupy on the edge server has not been determined. Next, the amount of resources that the power terminal device needs to occupy on the edge server will be analyzed.

[0041] Taking the second derivative of formula (23) gives: ; Due to the constraint of formula (20), , and is always greater than or equal to 0, that is . is a concave function. Therefore, there is a solution that minimizes the cost function for the offloading optimization problem of the power terminal device.

[0042] When the first derivative of the cost function of the power terminal device is equal to 0, the solution that minimizes can be obtained.

[0043] ; According to the constraint of formula (20), is always less than 0, and the amount of resources occupied by the power terminal device on the edge server , therefore, the optimal amount of resources occupied by the edge server is a positive value, that is ; wherein represents the amount of edge server resources that can reach the Nash equilibrium.

[0044] 3. Offloading Decision The optimal offloading scheme can be expressed as follows: When at that time ; ; ; ; When at that time .

[0045] wherein represents the amount of resources for optimal offloading, represents the amount of data for optimal offloading, represents the lower limit of the amount of edge server resources occupied, represents the upper limit of the amount of edge server resources occupied.

[0046] III. Design of Grid Intelligent Operation Computing Power Scheduling Weight Factors; In the mechanism described in this embodiment, the basic pricing function determines its weight factors and through the edge server. Thus, the optimal offloading decision of the power terminal device depends on the setting of these weight factors. In the context of dynamic games, if the configuration of the weight factors and is improper, it may lead to unreasonable evaluation of the cost function of the power terminal device, thereby affecting its offloading willingness and reducing the performance of the edge server. The core challenge in solving such problems lies in that the decision variables of the weight factors and evolve continuously in a complex environment, making it difficult to implement a fixed solution. Therefore, the research proposes to use a heuristic search algorithm to explore feasible solutions. However, the effectiveness of such algorithms is greatly affected by the setting of the search data granularity and range, and their execution time may be significantly extended. To address this issue, if the edge server has accumulated a large number of historical search results, a regression model can be constructed to approximate these solutions. In this way, the server can quickly provide suggestions based on the bidding mechanism to the power terminal device. The method proposed in this embodiment is based on supervised learning, and accurately determines the weight factors and by analyzing the accumulated data, in order to discover efficient offloading strategies.

[0047] This embodiment adopts a regression feedforward neural network under the supervised learning framework. The network design includes three hidden layers, consisting of 512, 256, and 128 neurons respectively. The ReLU function is used as the activation function for each layer. This embodiment generates two million data. Each data contains , , , , and . During the training process, =5 fold cross-validation method is applied to improve the training effect of the dataset. After training, the edge server will obtain a model for calculating the weight factors and . When receiving a new grid intelligent operation offloading request, the server can load the pre-trained model and quickly calculate the appropriate weight factors and .

[0048] This embodiment constructs a model based on dynamic game for each participant in the edge-side grid intelligent operation with heterogeneous computing power to address the problem of heterogeneous computing power scheduling in edge-side grid intelligent operations. The edge server and power terminal devices are regarded as game participants, where the edge server is the leader and the power terminal devices are the followers. The edge server determines the basic pricing function based on the grid intelligent operation-related information reported by the power terminal devices, such as local computing power resources, the amount of grid intelligent operation data, and the latency requirements of grid intelligent operations. The power terminal devices then determine the amount of data to be offloaded to the edge server and the computing power resources of the edge server to be occupied based on this basic pricing function, taking into account their own execution requirements and costs. In this process, the goals of both sides restrict and influence each other. The edge server aims to maximize its own utility, while the power terminal devices seek to minimize the execution cost of grid intelligent operations (including computing latency and resource usage costs). Through such a dynamic game process, the reasonable allocation and scheduling of edge-side computing power resources are achieved, the overall intelligent operation efficiency is improved, the situation of overload or idle waste of resources on the edge server is avoided, and at the same time, the execution latency is ensured to meet the requirements of grid intelligent operations.

[0049] This embodiment proposes an innovative pricing strategy that fully considers the real-time load situation of edge servers. The determination of the unit resource price is no longer determined by a single factor, but is closely related to the remaining computing power resources of the edge server (i.e., the workload status) and the computing power resources occupied by power terminal devices. When there is more remaining computing power in the edge server, the cost of power terminal devices occupying the same computing power resources is relatively low; conversely, if the edge server has a high load, the unit resource price of power terminal devices occupying computing power resources will increase accordingly. Such a pricing method encourages power terminal devices to be more cautious when offloading grid intelligent operations in choosing the required computing power resources, avoiding over-occupation that may cause server overload. At the same time, it also motivates edge servers to reasonably manage and allocate their limited computing power resources, improve resource utilization efficiency, and ensure that grid intelligent operations can run stably and efficiently under different load conditions.

[0050] In view of the fact that the parameters involved in the dynamic game process (such as the weight factors in the pricing strategy, etc.) have a key impact on the computing power scheduling effect, and these parameters will change continuously in the complex grid intelligent operation environment, this embodiment adopts a supervised learning-based method to determine these parameters. Specifically, a feedforward neural network is used to construct a network structure with multiple hidden layers (a three-layer hidden layer composed of 512, 256, and 128 hidden neurons respectively) and the activation function is ReLU. By collecting a large amount of historical data related to grid intelligent operations, including the computing power situation of different power terminal devices, the data characteristics of grid intelligent operations, the server load change situation, etc., training datasets and test datasets are generated. The K-fold cross-validation technique with K = 5 is used to train the neural network, enabling the edge server to quickly and accurately determine appropriate parameter values according to these trained models when facing new grid intelligent operation offloading requests, thereby realizing a more optimized computing power scheduling strategy, adapting to the changing needs in the grid intelligent operation scenario, and improving the adaptability and stability of the entire system.

[0051] In this embodiment, an optimization method for heterogeneous computing power scheduling of edge-grid intelligent operations based on dynamic game is designed. A simulation environment is constructed with reference to the characteristics of common edge-grid intelligent operations in the power grid to verify the effectiveness of the method. In the experiment, multiple scenarios of different arrival rates of edge-grid intelligent operations are set to simulate different workload conditions in the actual operation of the power grid. For each scenario, the performance of the proposed method and other traditional methods is compared and analyzed in multiple key performance indicators. These indicators include the average cost of power terminal devices, average execution delay, revenue of edge servers, and resource utilization rate of edge servers, etc. The experimental results show that under different arrival rates of edge-grid intelligent operations, the proposed method always performs excellently in terms of average execution delay, can effectively reduce the execution time of edge-grid intelligent operations, and meet the strict real-time requirements of edge-grid intelligent operations; at the same time, in terms of the resource utilization rate of edge servers, it can reasonably allocate resources according to the load conditions, avoid server overload, improve resource utilization efficiency, and strongly prove the effectiveness and superiority of the proposed method in heterogeneous computing power scheduling of edge-grid intelligent operations, providing reliable technical support for computing power management in the development of grid intelligence.

[0052] In practical applications, the optimization method for heterogeneous computing power scheduling of edge-grid intelligent operations in this embodiment can be widely applied in the power industry to improve the efficiency of edge-grid intelligent operations and achieve the rationality of computing power scheduling in scenarios where multiple entities share heterogeneous computing power resources.

[0053] For the heterogeneous computing power scheduling of edge-grid intelligent operations, this embodiment proposes an optimization method for heterogeneous computing power scheduling of edge-grid intelligent operations based on dynamic game according to the status information of servers and the status information of edge-grid intelligent operations, avoiding waste and bottlenecks of resources, and providing strong technical support for the efficient and stable operation of edge-grid intelligent operations in scenarios where multiple entities share heterogeneous computing power resources.

[0054] Embodiment 2 provides a device for optimizing heterogeneous computing power scheduling of edge-grid intelligent operations, including: A model construction module, which is used to set the edge server as the leader role in the game and the power terminal device as the follower role, and construct a dynamic game model for edge-grid intelligent operation computing power scheduling, where: The edge server determines a basic pricing function according to the information related to edge-grid intelligent operations reported by the power terminal device, The power terminal device determines the amount of data to be offloaded to the edge server and the amount of edge server resources required according to the basic pricing function, combined with its own execution requirements and cost considerations; A scheduling module, which is used to obtain the optimal offloading strategy based on the dynamic game process of the dynamic game model for edge-grid intelligent operation computing power scheduling, and realize the reasonable allocation and scheduling of edge-side computing power resources.

[0055] For the specific function implementation of each of the above modules, refer to the relevant content in the method of Embodiment 1, which will not be elaborated here.

[0056] Embodiment 3 provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps of the method described in any one of Embodiment 1.

[0057] Embodiment 4 provides a computer device, including: A memory for storing computer programs / instructions; A processor for executing the computer programs / instructions to implement the steps of the method described in any one of Embodiment 1.

[0058] Embodiment 5 provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by a processor, they implement the steps of the method described in any one of Embodiment 1.

[0059] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

[0060] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as methods, systems, or computer program products. Therefore, the present disclosure can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0062] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the function specified in one process or more processes and / or one block or more blocks of the process Figure 1 in one process or more processes and / or Figure 1 one block or more blocks of the block

[0063] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the function specified in one process or more processes and / or one block or more blocks of the process Figure 1 in one process or more processes and / or Figure 1 one block or more blocks of the block

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit the scope of its protection. Although the present disclosure has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present disclosure, those skilled in the art can still make various changes, modifications, or equivalent replacements to the specific implementation manners of the invention, but these changes, modifications, or equivalent replacements are all within the scope of protection of the pending claims of the disclosure.

Claims

1. A method for optimizing the scheduling of heterogeneous computing power for intelligent operations in edge power grids, characterized in that: include: The edge server is set as the leader role in the game, and the power terminal equipment is set as the follower role to build a dynamic game model for intelligent operation computing power scheduling of power grids, where: The edge server determines the basic pricing function based on the information related to the smart operation of the power grid reported by the power terminal equipment. The power terminal device determines the amount of data to be unloaded to the edge server and the amount of edge server resources to be occupied according to the basic pricing function and in combination with its own execution requirements and cost considerations; Based on the dynamic game process of the dynamic game model for scheduling computing power of intelligent power grid operations, the optimal unloading strategy is obtained to achieve reasonable allocation and scheduling of edge computing power resources.

2. The method for optimizing the scheduling of heterogeneous computing power for intelligent operations in edge power grids according to claim 1 is characterized in that: The calculation formula of the basic pricing function is as follows: ； in, represents the basic pricing function, Indicates the amount of resources occupied by smart grid operations, Indicates the remaining resources of the edge server. When the power grid is intelligent, it can unload and is the weight factor, which is a positive value and is determined by the edge server.

3. The method for optimizing the scheduling of heterogeneous computing power for intelligent operations in edge power grids according to claim 1 is characterized in that: The information related to the smart grid operation includes the amount of local computing resources, the amount of data of the smart grid operation, and the execution delay of the smart grid operation.

4. The method for optimizing the scheduling of heterogeneous computing power for intelligent operations in edge power grids according to claim 3 is characterized in that: The execution delay of the power grid smart operation is calculated as follows: ； in, Indicates the execution delay of the grid smart operation, Indicates the upload rate. Indicates the download rate. Indicates the size ratio before and after data processing, represents the amount of data sent to the edge server for computation. Indicates the amount of power grid intelligent operation data, Indicates the amount of resources consumed by the smart grid operation when processing each bit of data. Indicates the amount of resources occupied by smart grid operations, Indicates the amount of local computing resources.

5. The method for optimizing the scheduling of heterogeneous computing power for intelligent operations in edge power grids according to claim 4 is characterized in that: The dynamic game process based on the dynamic game model of the power grid intelligent operation computing power scheduling obtains the optimal unloading strategy, including: When the time for processing the grid smart operation locally in the power terminal device is equal to the total time for offloading the grid smart operation to the edge server for processing, the amount of data transmitted to the edge server is the balanced data. , the formula is as follows: ； The minimum cost function of the power terminal equipment is calculated based on the balance data. The calculation formula is as follows: ； in, represents the uninstall weight, Indicates the remaining resources of the edge server; The problem of optimizing the unloading of power terminal equipment The statement is as follows: ； right Find the second-order derivative and find the order The smallest solution, which is the amount of edge server resources that can achieve Nash equilibrium , the calculation formula is as follows: ； The optimal unloading strategy is expressed as follows: when hour, ； ； ； ； when hour, ; in represents the optimal amount of resources to offload, Indicates the optimal amount of data to be unloaded, Indicates the lower limit of the edge server resources occupied. The upper limit of the edge server resources occupied is shown. Represents the delay constraints of the power grid smart operation.

6. The method for optimizing the dispatching of heterogeneous computing power for intelligent operations in edge power grids according to claim 2 is characterized in that: The method also includes: using a supervised learning method to determine the weight factor in the basic pricing function and Specifically, a feedforward neural network is used to construct a network structure containing multiple hidden layers, and a large amount of historical data related to the smart operation of the power grid is collected for training, so that the edge server can determine the weight factor according to the trained model. and The value of .

7. A device for optimizing the dispatching of heterogeneous computing power for intelligent operations in edge power grids, characterized in that: include: The model building module is used to set the edge server as the leader role in the game and the power terminal equipment as the follower role to build a dynamic game model for intelligent operation computing power scheduling of the power grid, where: The edge server determines the basic pricing function based on the information related to the smart operation of the power grid reported by the power terminal equipment. The power terminal device determines the amount of data to be unloaded to the edge server and the amount of edge server resources to be occupied according to the basic pricing function and in combination with its own execution requirements and cost considerations; The scheduling module is used to obtain the optimal unloading strategy based on the dynamic game process of the dynamic game model for scheduling computing power of intelligent power grid operations, so as to realize the reasonable allocation and scheduling of terminal computing power resources.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: include: Memory, for storing computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images