Power grid voltage control task unloading and computing resource allocation method, equipment and medium

Abstract

The invention discloses a power grid voltage control task unloading and computing resource allocation method and device and a medium, and the method comprises the steps: setting a power grid voltage node partition model according to the number of nodes and the load condition, and solving the calculation time delay of a voltage control task of each partition according to the number of nodes and the calculation complexity of different partitions; calculating the time delay based on each partition and then solving the global task calculation time delay; after the global calculation time delay is solved, the energy consumption in the task calculation process is solved; after the energy consumption in the task calculation process is solved, modeling is carried out to solve time delay and energy consumption minimization; and finally, determining an objective function and constraint conditions based on time delay and energy consumption minimization, and solving by using a deep reinforcement learning algorithm to obtain a power grid voltage control task unloading and computing resource allocation strategy. According to the method, a power grid partitioning strategy is optimized, and calculation resources are reasonably allocated, so that the calculation time delay and the calculation energy consumption cost of a voltage control task are minimized, and the self-adaptability and the robustness of a power grid regulation and control system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system resource scheduling and optimization, and in particular to a method, device and medium for unloading power grid voltage control tasks and allocating computing resources. Background Art

[0002] As the global energy system evolves towards intelligence and digitalization, the automation and intelligence level of power grids continues to improve, and the construction of smart grids has become a key direction for the development of modern power systems. Smart grids integrate advanced power control technologies and information and communication technologies to achieve accurate perception, efficient calculation and optimized control of the operating status of power grids, thereby improving the safety, stability and economy of power grid operation. Among them, the voltage control system based on the edge computing architecture relies on distributed computing capabilities to effectively integrate various computing resources, achieve rapid response and precise execution of voltage control tasks, and provide technical support for the efficient operation of the power grid.

[0003] However, with the continuous expansion of the scale of power grids and the increase in the proportion of distributed energy access, the topology of the power grid has become more complex, and the traditional centralized voltage control mode faces many challenges such as concentrated computing load, increased processing delay, and limited system scalability. Especially in a multi-node distributed architecture, how to efficiently handle voltage control tasks and achieve optimal computing resource allocation under the constraints of limited computing resources has become a key problem affecting the efficiency of power grid voltage regulation.

[0004] At present, the power grid voltage control task mainly relies on centralized scheduling strategies, that is, the data of all nodes must be transmitted to the central server for calculation and decision-making. However, this approach is prone to increase computing delay and system response lag when facing dynamic load changes, network congestion or communication interruption, which in turn affects the real-time and stability of voltage control. In addition, since the voltage control task has the characteristics of high computational complexity and strict timeliness requirements, how to optimize the partitioning strategy of power grid nodes and reasonably allocate computing resources to minimize computing delay and computing energy consumption costs while ensuring control accuracy is the core issue of current smart grid computing optimization research.

[0005] In view of the above problems, it is urgent to explore a voltage control task offloading and computing resource optimization allocation method suitable for large-scale distributed power grid architecture, which can dynamically adjust the partitioning strategy so that computing resources can be reasonably allocated between different task units, thereby reducing the overall computing energy consumption while meeting the delay constraints. The breakthrough of this technology will significantly improve the adaptability, robustness and computing efficiency of smart grid voltage control, help the power grid control system develop in a more intelligent, efficient and low-carbon direction, and provide important support for the construction of a new energy Internet.

[0006] The above contents are only used to assist in understanding the technical solution of the present application, and do not constitute an admission that the above contents are the closest prior art. Summary of the invention

[0007] The main purpose of the present invention is to provide a method for unloading power grid voltage control tasks and allocating computing resources to solve the technical problems raised in the background technology.

[0008] The present invention adopts the following technical solutions to solve the above technical problems: A method for unloading power grid voltage control tasks and allocating computing resources, wherein a computer device performs the following steps: S1. According to the number of nodes and load conditions, the grid voltage node partition model is set, and the number of nodes and computational complexity of different partitions are considered, and the computational delay of the voltage control task of each partition is calculated; S2, based on S1, obtains the computation delay of each partition and then calculates the global task computation delay; S3. After the global computing delay is obtained, the energy consumption during the task computing process is obtained; S4. After finding the energy consumption in the task calculation process, modeling is performed to minimize the delay and energy consumption; S5. Finally, the objective function and constraints are determined based on minimization of delay and energy consumption, and the deep reinforcement learning algorithm is used to solve the grid voltage control task unloading and computing resource allocation strategy.

[0009] Furthermore, step S1 specifically includes: According to the number of nodes and load conditions, the nodes in the physical area are divided into several subsets, each of which is regarded as an independent task unit to independently calculate the corresponding voltage control task; Assume that there is The macro base station needs to partition the nodes. The number of nodes in ,have partitions, the partition strategy is expressed as: , (1) in, is a set of partitions, For the partition index.

[0010] Furthermore, step S1 further includes: Allocate computing resources according to the number of nodes and computing complexity of different partitions to minimize task processing latency and cost; Assume the computing power of the base station is , allocate computing resources to each partition ,satisfy ,The allocation strategy affects the task execution delay and computing energy consumption. The computing delay of each partition is expressed as: (2) in, For partition The computational complexity of the voltage control task.

[0011] Furthermore, step S2 specifically includes: Take the maximum latency in each partition as the global task calculation latency , expressed as: (3).

[0012] Further, step S3 specifically includes: During the task calculation process, the server provides computing resources for the partitioned tasks. The energy consumption associated with this process is The computational energy consumption is : (4) in, is the energy consumption factor.

[0013] Further, step S4 specifically includes: The problem of minimizing computational delay and energy consumption is stated as follows: (5) in, and is a weight factor, used to indicate the importance of latency and energy consumption; is the delay constraint of the global voltage control task; Indicates computing resource constraints. The sum of computing resources that the server assigns to each partition must not exceed its own computing resources. Indicates the node number constraint. The sum of the number of nodes in each partition is consistent with the summary number of nodes. represents the delay constraint. The completion delay of the global voltage control task must not exceed the maximum delay tolerance of the global voltage control; Represents a weight constraint, the sum of weight factors does not exceed 1.

[0014] Furthermore, the specific steps of step S5 include: 1) Initialize the power grid system parameters, including: power grid nodes, base station computing resource upper limit and task delay threshold; define and initialize the deep reinforcement learning model. In the deep reinforcement learning framework, each partition The state space is: (6) in, is the total number of nodes, For node partitioning, Calculate the total amount of resources for the macro base station, Provide feedback for historical decision making; The action space is: (7) in, Indicates adjusting the number of partition nodes. Indicates adjustment of computing resource allocation; Partition The reward function is: (8) in, is a weight parameter, representing the influence weight of computing delay and computing energy consumption; The policy network for decision-making computing resource allocation strategy is ; Initialize the global model and buffer ,The global model is used to store the aggregated global parameters, and the buffer is used to temporarily store the local model parameters collected from each partition; 2) After initializing the parameters of the power grid system, proximal policy optimization is used for learning. Specifically, the power grid control strategy is optimized based on the policy-evaluation architecture. The policy network is responsible for generating the optimal action, namely the power grid partitioning scheme and the computing resource allocation ratio: (9) Proximal strategy optimization goals: (10) in, is the strategy change ratio, Evaluate the superiority of the current action, To limit the scope of policy changes; The evaluation network is responsible for evaluating the value of the current state and guiding the strategy network to learn: (11) The evaluation network uses mean square error optimization: (12) in, ; 3) After learning with proximal strategy optimization, training and optimization are performed, including initializing the strategy network and evaluating network parameters, setting the initial grid partitioning scheme and computing resource allocation scheme, and Take Action , update grid partitioning and resource allocation; calculate rewards And store experience data, i.e. state, action, reward, next state; update the evaluation network, optimize the state value estimate; calculate the advantage estimate And update the policy network; repeat the above steps until the algorithm converges or reaches the upper limit of the number of training rounds.

[0015] On the other hand, the present invention further discloses a computer-readable storage medium, characterized in that a computer program is stored therein, and when the computer program is executed by a processor, the processor executes the steps of the above method.

[0016] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0017] It can be seen from the above technical solution that the present invention provides a method for unloading power grid voltage control tasks and allocating computing resources. The method aims to optimize the power grid partitioning strategy and reasonably allocate computing resources, thereby minimizing the computing delay and computing energy consumption cost of the voltage control task, and improving the adaptability and robustness of the power grid control system.

[0018] Specifically, the advantages of the present invention are as follows: 1. Real-time partition adjustment: Dynamically divide the power grid into zones through deep reinforcement learning, adjust the partition strategy according to real-time data such as renewable energy output fluctuations and load changes, and solve the rigidity problem of traditional static partitions.

[0019] 2. Elastic resource allocation: Based on task complexity and node load, computing resources are dynamically allocated to achieve “partition-resource” collaborative optimization and adapt to sudden load changes in seconds (such as electric vehicle fast charging and industrial impact loads).

[0020] 3. Multi-objective optimization: Minimize computing latency and energy consumption simultaneously to avoid resource waste or overload risks.

[0021] 4. Closed-loop control strategy: A closed loop is formed from data collection, task offloading to control feedback to improve the grid’s ability to resist interference under extreme fluctuations (such as sudden drops in wind and solar power, sudden changes in load).

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easy to understand through the following description. Of course, it is not necessary to achieve all of the advantages described above simultaneously for any product implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 System architecture of an embodiment of the present invention; Figure 2 Flowchart of the system of the present invention. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] In the embodiment, see Figure 1 to Figure 2 .

[0026] like Figure 1 As shown, the grid voltage control task unloading and computing resource allocation method proposed in the embodiment of the present invention is based on a grid voltage control system, wherein the grid voltage control system is composed of a base station and multiple distributed grid nodes, and adopts a closed-loop control strategy of partitioning-resource optimization-task unloading-control feedback to achieve efficient voltage regulation. The system includes the following key components: 1. Partition management module: According to the load of the grid nodes and the computing power of the server, the entire grid is divided into multiple partitions. Each partition is an independent task unit, and the base station coordinates computing resources for optimized calculation. The optimization goal of the partition strategy is to minimize the computing delay and computing energy consumption cost. The core optimization variables include the number of nodes in each partition and the load balance between partitions.

[0027] 2. Computing resource allocation module: The base station is responsible for allocating computing resources to each partition to ensure that the computing task is completed under the optimal conditions of latency constraints and energy consumption. The computing resource allocation strategy considers the number of partition nodes, task complexity, and computing power of each partition to dynamically adjust the resource allocation plan to avoid waste or overload of computing resources.

[0028] 3. Task offloading and execution module: The grid node transmits the collected voltage data to the base station. The base station generates the corresponding task calculation model according to the partition division, and uses deep reinforcement learning strategy to optimize the number of partition nodes and computing resource scheduling.

[0029] 4. Voltage control instruction generation and feedback module: After the base station completes the task calculation, it sends the optimized voltage control instruction to the control equipment in each partition, and the controller performs voltage regulation to achieve dynamic optimization of the grid voltage.

[0030] like Figure 2 As shown, the grid voltage control task unloading and computing resource allocation method proposed in the embodiment of the present invention specifically includes the following steps: Step 1: Node partition model According to the number of nodes and load conditions, the nodes in the physical area are divided into several subsets, and each subset is regarded as an independent task unit to independently calculate the corresponding voltage control task. The macro base station needs to partition the nodes. The number of nodes in ,have partitions, the partition strategy can be expressed as: , (1) in, is a set of partitions, For the partition index.

[0031] Step 2: Calculate the delay model According to the number of nodes and computational complexity of different partitions, computing resources are reasonably allocated to minimize task processing delay and cost. Assume that the computing power of the base station is , allocate computing resources to each partition ,(satisfy ). The allocation strategy affects the task execution delay and computing energy consumption. The computing delay of each partition can be expressed as: (2) in, For partition The computational complexity of the voltage control task.

[0032] Step 3: Global delay model Take the maximum latency in each partition as the global task calculation latency , which can be expressed as: (3) Step 4: Calculate the energy consumption model During the task calculation process, the server provides computing resources for the partitioned tasks. The energy consumption associated with this process is The computational energy consumption is : (4) in, is the energy consumption factor.

[0033] Step 5: Determine the objective function and constraints The present invention jointly optimizes the number of node partitions and the allocation of computing resources of the server to each partition. The problem of minimizing the computing delay and energy consumption of the strategy of the present invention is expressed as follows: (5) in, and is a weight factor, used to indicate the importance of latency and energy consumption; is the delay constraint of the global voltage control task. Indicates computing resource constraints. The sum of computing resources that the server assigns to each partition must not exceed its own computing resources. Indicates the node number constraint. The sum of the number of nodes in each partition is consistent with the summary number of nodes. represents the delay constraint. The completion delay of the global voltage control task must not exceed the maximum delay tolerance of the global voltage control; Represents a weight constraint, the sum of weight factors does not exceed 1.

[0034] Step 6: Use deep reinforcement learning algorithm to solve Under the given constraint framework, this paper uses a deep reinforcement learning algorithm to solve the objective function, aiming to discover the optimal number of partition nodes and the allocation of computing resources. The specific steps are as follows: System initialization defines the grid parameters and model architecture, providing the basic rules for state space, action space and reward calculation for decision optimization. Initialize the grid system parameters, including: grid nodes, base station computing resource upper limit and task delay threshold. Define and initialize the deep reinforcement learning model. In the deep reinforcement learning framework, each partition The state space is: (6) in, is the total number of nodes, For node partitioning, Calculate the total amount of resources for the macro base station, Provide feedback for historical decision making.

[0035] The action space is: (7) in, Indicates adjusting the number of partition nodes. Indicates adjusting computing resource allocation.

[0036] Partition The reward function is: (8) in, is a weight parameter, representing the weight of the computational delay and computational energy consumption. The goal of the present invention is to minimize the delay and energy consumption, so it takes a negative value.

[0037] The policy network for decision-making computing resource allocation strategy is . Initialize the global model and buffer ,The global model is used to store the aggregated global parameters, and the buffer is used to temporarily store the local model parameters collected from each partition.

[0038] The actions generated by the decision optimization strategy network are used in the actual power grid environment, and the generated interactive data is fed back to the training process to form a "decision-execution-feedback" closed loop. Specifically, proximal policy optimization (PPO) is used for learning, and the power grid control strategy is optimized based on the strategy-evaluation architecture. The strategy network is responsible for generating the optimal action, that is, the power grid partitioning scheme and the computing resource allocation ratio: (9) Proximal strategy optimization goals: (10) in, is the strategy change ratio, Evaluate the superiority of the current action, Limit the scope of policy changes.

[0039] The evaluation network is responsible for evaluating the value of the current state and guiding the strategy network to learn: (11) The evaluation network uses mean square error optimization: (12) in, .

[0040] Training and Optimization The policy network and the evaluation network are updated alternately during training - the policy network optimizes the action generation strategy, and the evaluation network optimizes the state value estimate. Tight coupling jointly drives the convergence of the global optimal solution for multiple objectives (delay, energy consumption). By periodically synchronizing the global model and continuously collecting real-time data, the system can adapt to grid load fluctuations and realize dynamic adjustment of partitioning strategies and resource allocation.

[0041] Specifically, it is to initialize the strategy network and evaluate network parameters, set the initial grid partitioning scheme and computing resource allocation scheme. 2. Based on the current state Take Action , update grid partitioning and resource allocation. 3. Calculate rewards And store the experience data (state, action, reward, next state). 4. Update the evaluation network and optimize the state value estimate. 5. Calculate the advantage estimate And update the policy network. 6. Repeat steps 2-5 until the algorithm converges or reaches the upper limit of the number of training rounds.

[0042] The above system initialization parameters are directly used as inputs and constraints for decision optimization; the actions generated by decision optimization are used for environmental interaction to generate training data; the training process uses the interaction data to update the strategy and evaluation network to form a closed-loop feedback; and finally the converged strategy feeds back to the power grid system to achieve dynamic control goals.

[0043] In summary, the advantages of the embodiments of the present invention are as follows: 1. Real-time partition adjustment: Dynamically divide the power grid into zones through deep reinforcement learning, adjust the partition strategy according to real-time data such as renewable energy output fluctuations and load changes, and solve the rigid problem of traditional static partitions. 2. Elastic resource allocation: Based on task complexity and node load, dynamically allocate computing resources to achieve "partition-resource" collaborative optimization and adapt to sudden load changes in seconds (such as electric vehicle fast charging and industrial impact load). 3. Multi-objective optimization: Minimize computing latency and energy consumption at the same time to avoid resource waste or overload risks. 4. Closed-loop control strategy: From data collection, task offloading to control feedback, a closed loop is formed to improve the anti-interference ability of the power grid under extreme fluctuations (such as sudden drops in wind and solar power, sudden load changes). Based on the above theory, take an application scenario as an example to describe the entire practice process: Renewable energy generation such as wind and solar power has strong volatility, which leads to frequent fluctuations in grid voltage. Traditional static partition strategies are difficult to adapt to dynamic load changes, which can easily cause local voltage limit violations or imbalanced resource allocation. The peak-to-valley difference in the commercial area during the day is as high as 3:1. Traditional centralized control has communication delays, and the idle rate of computing resources is high when the load is light at night. The impact load of the steel plant causes voltage flicker, the local controller has insufficient computing power, and outsourced cloud computing has data security risks. Therefore, the present invention is suitable for actual scenarios such as urban high-density distribution networks and industrial park microgrids.

[0044] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0045] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0046] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the grid voltage control task unloading and computing resource allocation methods in the above embodiments.

[0047] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.

[0048] The embodiment of the present application also provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. Memory, used to store computer programs; The processor is used to implement the above-mentioned grid voltage control task unloading and computing resource allocation method when executing the program stored in the memory.

[0049] The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc.

[0050] The communication interface is used for communication between the above electronic device and other devices.

[0051] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0052] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0053] It should also be noted that electronic devices also include terminal devices, which can also be called terminals, user equipment (UE), mobile stations (MS), mobile terminals (MT), etc. Terminal devices can be mobile phones, smart TVs, wearable devices, tablet computers (Pad), computers with wireless transceiver functions, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc. The embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal devices.

[0054] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0056] In addition, it should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0057] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing in the full text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or schemes that A and B meet at the same time. In addition, in the embodiments of the present invention, "multiple" refers to more than two. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A method for unloading power grid voltage control tasks and allocating computing resources, characterized in that: S1. According to the number of nodes and load conditions, the grid voltage node partition model is set, and the number of nodes and computational complexity of different partitions are considered, and the computational delay of the voltage control task of each partition is calculated; S2, based on S1, obtains the computation delay of each partition and then calculates the global task computation delay; S3. After the global computing delay is obtained, the energy consumption during the task computing process is obtained; S4. After finding the energy consumption in the task calculation process, modeling is performed to minimize the delay and energy consumption; S5. Finally, the objective function and constraints are determined based on the minimization of delay and energy consumption, and the deep reinforcement learning algorithm is used to solve the grid voltage control task unloading and computing resource allocation strategy; Step S1 specifically includes: According to the number of nodes and load conditions, the nodes in the physical area are divided into several subsets, each of which is regarded as an independent task unit to independently calculate the corresponding voltage control task; Assume that there is The macro base station needs to partition the nodes. The number of nodes in ,have partitions, the partition strategy is expressed as: ， （1） in, is a set of partitions, For the partition index.

2. The method for unloading grid voltage control tasks and allocating computing resources according to claim 1, characterized in that: Step S1 also includes, Allocate computing resources according to the number of nodes and computing complexity of different partitions to minimize task processing latency and cost; Assume the computing power of the base station is , allocate computing resources to each partition ,satisfy ,The allocation strategy affects the task execution delay and computing energy consumption. The computing delay of each partition is expressed as: （2） in, For partition The computational complexity of the voltage control task.

3. The method for unloading grid voltage control tasks and allocating computing resources according to claim 2, characterized in that: Step S2 specifically includes: Take the maximum latency in each partition as the global task calculation latency , expressed as: （3）。 4. The method for unloading grid voltage control tasks and allocating computing resources as claimed in claim 3, characterized in that: Step S3 specifically includes: During the task calculation process, the server provides computing resources for the partitioned tasks. The energy consumption associated with this process is The computational energy consumption is : （4） in, is the energy consumption factor.

5. The method for unloading power grid voltage control tasks and allocating computing resources as claimed in claim 4, characterized in that: Step S4 specifically includes: The problem of minimizing computational delay and energy consumption is stated as follows: （5） in, and is a weight factor, used to indicate the importance of latency and energy consumption; is the delay constraint of the global voltage control task; Indicates computing resource constraints. The sum of computing resources that the server assigns to each partition must not exceed its own computing resources. Indicates the node number constraint. The sum of the number of nodes in each partition is consistent with the summary number of nodes. represents the delay constraint. The completion delay of the global voltage control task must not exceed the maximum delay tolerance of the global voltage control; Represents a weight constraint, the sum of weight factors does not exceed 1.

6. The method for unloading grid voltage control tasks and allocating computing resources as claimed in claim 5, characterized in that: The specific steps of step S5 include: 1) Initialize the power grid system parameters, including: power grid nodes, base station computing resource upper limit and task delay threshold; define and initialize the deep reinforcement learning model. In the deep reinforcement learning framework, each partition The state space is: （6） in, is the total number of nodes, For node partitioning, Calculate the total amount of resources for the macro base station, Provide feedback for historical decision making; The action space is: （7） in, Indicates adjusting the number of partition nodes. Indicates adjustment of computing resource allocation; Partition The reward function is: （8） in, is a weight parameter, representing the influence weight of computing delay and computing energy consumption; The policy network for decision-making computing resource allocation strategy is ; Initialize the global model and buffer ,The global model is used to store the aggregated global parameters, and the buffer is used to temporarily store the local model parameters collected from each partition; 2) After initializing the parameters of the power grid system, proximal policy optimization is used for learning. Specifically, the power grid control strategy is optimized based on the policy-evaluation architecture. The policy network is responsible for generating the optimal action, namely the power grid partitioning scheme and the computing resource allocation ratio: （9） Proximal strategy optimization goals: （10） in, is the strategy change ratio, Evaluate the superiority of the current action, To limit the scope of policy changes; The evaluation network is responsible for evaluating the value of the current state and guiding the strategy network to learn: （11） The evaluation network uses mean square error optimization: （12） in, ; 3) After learning with proximal strategy optimization, training and optimization are performed, including initializing the strategy network and evaluating network parameters, setting the initial grid partitioning scheme and computing resource allocation scheme, and Take Action , update grid partitioning and resource allocation; calculate rewards And store experience data, i.e. state, action, reward, next state; update the evaluation network, optimize the state value estimate; calculate the advantage estimate And update the policy network; repeat the above steps until the algorithm converges or reaches the upper limit of the number of training rounds.

7. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to execute the steps of the method according to any one of claims 1 to 6.

8. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 6.

Images