Power equipment edge gateway computing power dynamic distribution method

Abstract

The invention is applicable to the technical field of computing power distribution, and provides a power equipment edge gateway computing power dynamic distribution method, which comprises the following steps of: determining a system state representation vector based on multi-source data of power equipment; the system state representation vector comprises a real-time computing power occupancy rate, a task queue length, a network bandwidth utilization rate and an equipment energy consumption index; constructing target parameters and constraint conditions of the calculation power dynamic distribution model; taking the system state representation vector as input, solving the computing power dynamic allocation model through a reinforcement learning algorithm, and generating a task scheduling path and a resource allocation proportion; and optimizing a preset hierarchical buffer queue based on the task scheduling path and the resource allocation proportion, and generating a target dynamic allocation strategy. According to the method, the high-real-time task response speed and the resource utilization rate are improved, energy consumption is reduced, meanwhile, backlog tasks are efficiently processed in the idle period through an event-driven mechanism, and system load balance is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of computing power distribution technology, and in particular to a method for dynamically allocating computing power of an edge gateway of an electric power device. Background Art

[0002] With the rapid development of smart grids and the Industrial Internet of Things, edge gateways for power equipment must process massive amounts of heterogeneous data while meeting the requirements of high real-time performance, low energy consumption, and high reliability. However, edge gateways have limited computing resources, and power equipment is often deployed in complex and changing network environments, leading to problems such as fluctuating network bandwidth, unbalanced task loads, and high device heterogeneity.

[0003] In existing technologies, dynamic allocation of computing power primarily relies on static priority scheduling or heuristic optimization algorithms. While static priority scheduling is based on preset rules and is simple to implement, it lacks the ability to dynamically detect changes in system status, which can easily lead to resource contention or the preemption of low-priority tasks in high-load scenarios. While heuristic algorithms can partially optimize resource allocation, their high computational complexity and slow convergence make them difficult to meet the real-time requirements of edge gateways. Furthermore, existing methods often focus on a single approach and fail to effectively integrate multi-source data to drive dynamic decision-making, resulting in allocation strategies that are insufficiently adaptable to real-world scenarios. Summary of the Invention

[0004] This application provides a method for dynamically allocating computing power of an edge gateway of an electric power equipment, which is used to solve the problem of insufficient adaptability of the allocation strategy to the actual scenario.

[0005] This application provides a method for dynamically allocating computing power of an edge gateway of an electric power device, including:

[0006] Determine a system status characterization vector based on multi-source data of power equipment; the system status characterization vector includes real-time computing power utilization, task queue length, network bandwidth utilization, and equipment energy consumption indicators;

[0007] Construct a dynamic computing power allocation model, where the target parameters include minimizing task response latency, minimizing data processing costs, minimizing energy consumption, and maximizing resource utilization; the constraints include ensuring that computing power resource usage does not exceed a preset upper limit, task response latency is below a real-time threshold, and energy consumption is below the device energy consumption limit;

[0008] Taking the system state representation vector as input, solving the computing power dynamic allocation model through a reinforcement learning algorithm to generate a task scheduling path and resource allocation ratio;

[0009] The preset hierarchical buffer queue is optimized based on the task scheduling path and resource allocation ratio to generate a target dynamic allocation strategy.

[0010] Furthermore, the system state characterization vector is determined based on the multi-source data of the power equipment; the system state characterization vector includes real-time computing power occupancy rate, task queue length, network bandwidth utilization rate and equipment energy consumption index, including:

[0011] The edge gateway collects sensor data, network topology status, task load characteristics, and environmental parameters of power equipment and pre-processes the collected data.

[0012] The preprocessed data is fused and mapped into a multi-dimensional vector to generate a system state representation vector.

[0013] Furthermore, the target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption and maximizing resource utilization, including:

[0014] Minimize task response delay T delay The expression:

[0015]

[0016] in: is the set of all tasks, K is the total number of tasks, D k is the data volume of the kth task, f CPU Calculate the CPU frequency of the edge gateway, B link is the network link bandwidth, η net is the network congestion factor, Q queue is the task queue length, μ process is the task processing rate, ω k is the task priority weight, satisfying

[0017] Furthermore, the target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption, and maximizing resource utilization, and also include:

[0018] Minimize data processing cost C processing The expression:

[0019]

[0020] in: is the subset of tasks that need to be processed. For the subset of tasks that require data transmission, is the subset of tasks that need memory storage, c CPU 、c mem and c trans They are CPU computing power unit price, memory storage unit price and data transmission unit price respectively. and are the CPU time and memory capacity occupied by the kth task, D k is the data transmission amount of the kth task.

[0021] Furthermore, the target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption, and maximizing resource utilization, and also include:

[0022] Minimize energy consumption E consumption The expression:

[0023]

[0024] Where: T is the total number of running time windows, P CPU 、P net and P mem They are CPU computing power, network module power and memory power respectively. and are the CPU computing, data transmission and memory usage in time window t respectively.

[0025] Furthermore, the target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption, and maximizing resource utilization, and also include:

[0026] Maximize resource utilization U resource The expression:

[0027]

[0028] Where: γ CPU , γ mem and γ bw are the weight coefficients of CPU, memory and bandwidth respectively, T total 、S total and B total They are total CPU time, total memory and total bandwidth capacity, is the effective bandwidth utilization of the kth task.

[0029] Furthermore, the constraints include computing resource usage not exceeding a preset upper limit, task response latency being lower than a real-time threshold, and energy consumption being lower than a device energy consumption limit, including:

[0030] The computing power resource usage does not exceed the preset upper limit:

[0031]

[0032] Where: T max 、S max and B max These are the preset upper limits for CPU, memory, and bandwidth respectively;

[0033] The task response latency is lower than the real-time threshold:

[0034]

[0035] Where: T threshold is the real-time threshold;

[0036]

[0037] Where: E limit The upper limit of device energy consumption.

[0038] Furthermore, the system state representation vector is used as input, and the computing power dynamic allocation model is solved by a reinforcement learning algorithm to generate a task scheduling path and a resource allocation ratio, including:

[0039] The reinforcement learning model includes a state space, an action space and a reward function; wherein the state space is composed of the system state representation vector, including real-time computing power occupancy, task queue length, network bandwidth utilization, device energy consumption indicators and task priority weights; the action space includes discrete task scheduling path decisions and continuous resource allocation ratios.

[0040] Furthermore, the system state representation vector is used as input, and the computing power dynamic allocation model is solved by a reinforcement learning algorithm to generate a task scheduling path and a resource allocation ratio, further comprising:

[0041] Input the current system state representation vector into a pre-trained deep reinforcement learning policy network, and output task scheduling path selection and resource allocation ratio decision; the task scheduling path is selected from a predefined set of candidate paths, including low-latency paths and high-throughput paths; the resource allocation ratio decision includes the allocation weights of CPU, memory, and bandwidth;

[0042] If the selected scheduling path is a low-latency path, high-real-time tasks will be assigned to the low-latency path first, and sufficient network bandwidth resources will be reserved; if the selected scheduling path is a high-throughput path, routine monitoring tasks will be assigned to the high-throughput path, and computing and storage resources will be dynamically allocated according to the resource allocation weights;

[0043] Based on the resource allocation weight, the CPU computing power, memory capacity and network bandwidth resources of the edge gateway are divided proportionally;

[0044] Based on the actual task execution delay, resource utilization and energy consumption data, the reinforcement learning reward value is calculated, and the subsequent decision-making is optimized through the policy network parameter update mechanism.

[0045] Furthermore, the preset hierarchical buffer queue is optimized based on the task scheduling path and resource allocation ratio to generate a target dynamic allocation strategy, including:

[0046] Based on the CPU and bandwidth weights in the resource allocation ratio, the scheduling frequency of high-real-time tasks in the priority queue is increased, where the high-real-time tasks are tasks whose response delay threshold is lower than a first preset value, and the scheduling weight of non-urgent tasks in the FIFO queue is reduced; if it is detected that the actual response delay of the high-real-time task reaches a second preset value, the capacity upper limit of the priority queue is increased, and some resources are preempted from the FIFO queue;

[0047] The event-driven mechanism is triggered during the system idle period, and the backlog tasks in the FIFO queue are dynamically pulled for batch processing based on the currently available computing resources;

[0048] The optimized queue scheduling rules and resource allocation ratio are determined as the target dynamic allocation strategy.

[0049] It can be seen from the above technical solutions that this application has the following advantages:

[0050] This application uses multi-source data fusion to perceive system status in real time, constructing a multi-objective optimization model to minimize latency, cost, energy consumption, and maximize resource utilization. It also dynamically generates task scheduling paths and resource allocation ratios based on a reinforcement learning algorithm. By optimizing the scheduling rules of hierarchical buffer queues, it prioritizes resource requirements for high-real-time tasks and utilizes an event-driven mechanism to batch process backlog tasks during idle periods, achieving efficient and flexible allocation of computing resources. This invention effectively improves the response speed and resource utilization of edge gateways, providing an adaptive, low-latency edge computing solution for scenarios such as smart grids and the Industrial Internet of Things. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The figure is a flow chart of an embodiment of a method for dynamically allocating computing power of an edge gateway of an electric power device in the present invention. DETAILED DESCRIPTION

[0052] The terms "first," "second," "third," "fourth," etc. (if any) in the specification and claims of the present application and in the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential sequence. It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to," and any variations thereof, are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or apparatus.

[0053] Example 1

[0054] The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, without specific limitation. The following will introduce the method for dynamically allocating computing power of the edge gateway of the power equipment in this application from the perspective of system implementation. Figure 1 , the method provided in the embodiment of the present application includes the following steps:

[0055] S11. Determine a system status representation vector based on multi-source data from power equipment; the system status representation vector includes real-time computing power utilization, task queue length, network bandwidth utilization, and equipment energy consumption indicators;

[0056] 1. Collect sensor data, network topology status, task load characteristics, and environmental parameters of power equipment through the edge gateway, and pre-process the collected data;

[0057] 2. Fuse the preprocessed data and map the fused data into a multi-dimensional vector to generate a system state representation vector.

[0058] Specifically, the edge gateway collects sensor data of power equipment in real time, such as temperature, voltage, current, network topology status, such as node connection relationship and link bandwidth, task load characteristics, such as task type, data volume, priority, and environmental parameters, and performs dynamic timestamp alignment and missing value interpolation preprocessing on multi-source heterogeneous data; based on the task priority weighting strategy and the dynamic fusion rules of network status, the preprocessed data is feature mapped according to the four core dimensions of real-time computing power occupancy, task queue length, network bandwidth utilization and equipment energy consumption indicators to generate a unified multi-dimensional system state representation vector, providing high-precision, low-latency global state input for the computing power dynamic allocation model.

[0059] S12. Build a dynamic computing power allocation model, where the target parameters include minimizing task response latency, minimizing data processing cost, minimizing energy consumption, and maximizing resource utilization; the constraints include ensuring that computing power resource usage does not exceed a preset upper limit, task response latency is below a real-time threshold, and energy consumption is below the device energy consumption limit;

[0060] Minimize task response delay T delay The expression:

[0061]

[0062] in: is the set of all tasks, K is the total number of tasks, D k is the data volume of the kth task, f CPU Calculate the CPU frequency of the edge gateway, B link is the network link bandwidth, η net is the network congestion factor, Q queue is the task queue length, μ process is the task processing rate, ω k is the task priority weight, satisfying

[0063] Minimize data processing cost C processing The expression:

[0064]

[0065] in: is the subset of tasks that need to be processed. For the subset of tasks that require data transmission, is the subset of tasks that need memory storage, c CPU 、c mem and c trans They are CPU computing power unit price, memory storage unit price and data transmission unit price respectively. and are the CPU time and memory capacity occupied by the kth task, D k is the data transmission amount of the kth task.

[0066] Minimize energy consumption E consumption The expression:

[0067]

[0068] Where: T is the total number of running time windows, P CPU 、P net and P mem They are CPU computing power, network module power and memory power respectively. and are the CPU computing, data transmission and memory usage in time window t respectively.

[0069] Maximize resource utilization U resource The expression:

[0070]

[0071] Where: γ CPU , γ mem and γ bw are the weight coefficients of CPU, memory and bandwidth respectively, T total 、S total and B total They are total CPU time, total memory and total bandwidth capacity, is the effective bandwidth utilization of the kth task.

[0072] Based on the expressions of the above target parameters, the objective function F of the dynamic computing power allocation model is obtained total for:

[0073] F total =λ1·T delay +λ2·C processing +λ3·E consumption ―λ4·U resource

[0074] Among them: λ1, λ2, λ3 and λ4 are weight coefficients used to balance the priority of each target.

[0075] The constraints are as follows:

[0076] The computing power resource usage does not exceed the preset upper limit:

[0077]

[0078] Where: T max 、S max and B max These are the preset upper limits for CPU, memory, and bandwidth respectively;

[0079] The task response latency is lower than the real-time threshold:

[0080]

[0081] Where: T threshold is the real-time threshold;

[0082]

[0083] Where: E limit The upper limit of device energy consumption.

[0084] S13. Using the system state representation vector as input, the dynamic computing power allocation model is solved by a reinforcement learning algorithm to generate a task scheduling path and resource allocation ratio;

[0085] Specifically, the reinforcement learning model includes state space, action space and reward function; among them, the state space is composed of the system state representation vector, including real-time computing power occupancy rate, task queue length, network bandwidth utilization, equipment energy consumption index and task priority weight; the action space includes discrete task scheduling path decision and continuous resource allocation ratio.

[0086] State space definition: The system state representation vector includes the following dimensions: real-time computing power utilization, task queue length, network bandwidth utilization, device energy consumption index and task priority weight, which are dynamically updated through preset rules. Action space definition: Discrete actions are selected from the predefined candidate path set P = {P low―latency ,P high―throughput}Select the scheduling path, P low―latency It is a low-latency path, suitable for high-real-time tasks (such as fault diagnosis), and requires a reserved bandwidth of ≥ 20%; P high―throughput It is a high-throughput path suitable for batch data processing tasks (such as log analysis) and allows delay tolerance. The continuous action is a resource allocation ratio vector.

[0087] Reward function:

[0088] R

[0089] = -λ1·T delay ―λ2·C processing ―λ3·E consumption +λ4·U resource ―λ5(U resource ―U max )

[0090] Among them: λ5 is the penalty coefficient, U max The upper limit of resource utilization.

[0091] Policy network training and decision generation include the following steps:

[0092] 1. Input the current system state representation vector into a pre-trained deep reinforcement learning policy network, which outputs task scheduling path selection and resource allocation ratio decisions. The task scheduling path is selected from a predefined set of candidate paths, including low-latency paths and high-throughput paths. The resource allocation ratio decision includes the allocation weights of CPU, memory, and bandwidth.

[0093] 2. If the selected scheduling path is a low-latency path, high-real-time tasks are preferentially assigned to the low-latency path, and sufficient network bandwidth resources are reserved. If the selected scheduling path is a high-throughput path, routine monitoring tasks are assigned to the high-throughput path, and computing and storage resources are dynamically allocated according to the resource allocation weights.

[0094] 2. Based on the resource allocation weights, the CPU computing power, memory capacity, and network bandwidth resources of the edge gateway are divided proportionally;

[0095] 3. Calculate the reinforcement learning reward value based on the actual task execution latency, resource utilization, and energy consumption data, and optimize subsequent decisions through the policy network parameter update mechanism.

[0096] The deep reinforcement learning model architecture adopts the deep deterministic policy gradient (DDPG) algorithm, which includes a policy network (Actor) and a value network (Critic). The policy network input is the state vector (system state representation vector), and the action generation is the policy network output scheduling path and resource allocation ratio. If P is selected low―latency , assign high real-time tasks to this path and reserve a bandwidth greater than 0.2; if P is selected high―throughput , allocate computing resources according to the CPU allocation ratio and storage allocation ratio, and allow task queues to be temporarily stored. Finally, calculate T based on the actual execution data of the task delay 、C processing 、E consumption and U resource , calculate the reward value R, and update the policy network and value network parameters through the temporal difference (TD) error.

[0097] The above steps use the reinforcement learning model to perceive the system status in real time, combine multi-objective optimization and constraints, dynamically generate task scheduling paths and resource allocation ratios, realize flexible computing power allocation of the edge gateway of power equipment, and significantly improve real-time performance, resource utilization and energy efficiency.

[0098] S14. Optimize the preset hierarchical buffer queue based on the task scheduling path and resource allocation ratio, and generate a target dynamic allocation strategy.

[0099] 1. Based on the CPU and bandwidth weights in the resource allocation ratio, the scheduling frequency of high-real-time tasks in the priority queue is increased. High-real-time tasks are tasks whose response latency threshold is lower than a first preset value, and the scheduling weight of non-urgent tasks in the FIFO queue is reduced. If the actual response latency of the high-real-time task reaches a second preset value, the capacity limit of the priority queue is increased, and some resources are preempted from the FIFO queue.

[0100] Task classification standards: High real-time tasks are defined as tasks with a response delay threshold lower than the first preset value, which is set at 100ms, such as equipment fault diagnosis and real-time control instructions. Non-urgent tasks (routine monitoring, log backup, etc.) do not have strict delay constraints and are stored in the FIFO queue.

[0101] Dynamic scheduling weight adjustment: Based on the CPU and bandwidth weights in the resource allocation ratio, the priority queue's scheduling frequency is increased proportionally. For example, if the CPU weight is 0.6, the priority queue's scheduling frequency is increased to 60% of the total resources. At the same time, the FIFO queue's scheduling weight is limited to 1-CPU weight to ensure that non-urgent tasks do not preempt high-time tasks.

[0102] Resource preemption mechanism: When it is detected that the actual response delay of a high-real-time task reaches the second preset value, which is 90ms, it is judged as "close to the risk of exceeding the limit"; the capacity upper limit of the priority queue is temporarily expanded by 20% (for example, from 50 tasks to 60 tasks), and an equal proportion of resources are preempted from the FIFO queue (such as releasing 10% of the FIFO queue capacity); by modifying the resource allocation table of the queue manager, some computing nodes and bandwidth resources of the FIFO queue are rebound to the priority queue.

[0103] 2. Trigger an event-driven mechanism during system idle periods to dynamically pull backlogged tasks from the FIFO queue for batch processing based on currently available computing resources;

[0104] Idle judgment criteria: Batch processing is triggered when the system meets the following conditions: CPU utilization is less than or equal to 20%; network bandwidth utilization is less than or equal to 15%; task queue length is greater than or equal to 50. The timer module of the edge gateway periodically detects the system status. If the idle condition is met for three consecutive detection cycles (30 seconds per cycle), the batch processing event is triggered. : The processing capacity of each batch is dynamically determined based on the current available computing power resources. For example, if the available CPU computing power is 4 cores, the number of tasks processed in a single batch is 10 tasks per core. Real-time monitoring of resource usage during processing. If U resource When the preset upper limit is approached, processing is immediately suspended and some resources are released.

[0105] 3. Determine the optimized queue scheduling rules and resource allocation ratio as the target dynamic allocation strategy.

[0106] The optimized queue scheduling rules (such as the priority queue expansion ratio and FIFO queue weight limit) and resource allocation ratio are encapsulated into a policy file in JSON format. A version number and timestamp are added to each generated policy to support historical policy backtracking and rapid rollback. The policy file is sent to the task distributor, resource scheduler, and queue manager in real time through the communication module of the edge gateway. Task latency, resource occupancy, and energy consumption are collected in real time. If it is detected that the value exceeds the upper limit, an alarm is triggered and the policy rollback process is started. If the resource or latency exceeds for two consecutive monitoring cycles (1 minute per cycle), it will automatically switch to the last valid policy version and re-call the reinforcement learning model to generate a revised policy.

[0107] This step achieves elastic optimization of hierarchical buffer queues by dynamically adjusting queue scheduling weights, resource preemption mechanisms, and batch processing during idle periods, combined with policy encapsulation and closed-loop monitoring.

[0108] The above-described embodiment uses multi-source data fusion to perceive system status in real time, constructs a multi-objective optimization model, and combines reinforcement learning to dynamically generate task scheduling paths and resource allocation ratios, optimizing the scheduling rules for hierarchical buffer queues. This solution significantly improves the response speed of high-real-time tasks, increases resource utilization, and reduces energy consumption. It also efficiently processes backlogged tasks during idle periods through an event-driven mechanism, ensuring system load balancing. Its adaptive adjustment capabilities and closed-loop feedback mechanism ensure flexible and robust computing power allocation in complex environments, providing an efficient and low-power edge computing solution for high-real-time scenarios such as smart grids.

[0109] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0110] In the embodiments provided herein, it should be understood that the division of units is merely a logical functional division. In actual implementation, other division methods may be employed, such as combining multiple units into one unit, splitting a unit into multiple units, or ignoring certain features. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically as a separate unit, or two or more units may be integrated into a single unit. These integrated units may be implemented in either hardware or software functional units.

[0111] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0112] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for dynamically allocating computing power of an edge gateway of an electric power equipment, characterized in that: include: Determine the system state representation vector based on multi-source data of power equipment; The system status characterization vector includes real-time computing power occupancy rate, task queue length, network bandwidth utilization rate and equipment energy consumption index; Construct a dynamic computing power allocation model, where the target parameters include minimizing task response latency, minimizing data processing costs, minimizing energy consumption, and maximizing resource utilization; the constraints include ensuring that computing power resource usage does not exceed a preset upper limit, task response latency is below a real-time threshold, and energy consumption is below the device energy consumption limit; Taking the system state representation vector as input, solving the computing power dynamic allocation model through a reinforcement learning algorithm to generate a task scheduling path and resource allocation ratio; The preset hierarchical buffer queue is optimized based on the task scheduling path and resource allocation ratio to generate a target dynamic allocation strategy.

2. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 1, characterized in that: The system state characterization vector is determined based on multi-source data of the power equipment; the system state characterization vector includes real-time computing power occupancy rate, task queue length, network bandwidth utilization rate and equipment energy consumption index, including: The edge gateway collects sensor data, network topology status, task load characteristics and environmental parameters of power equipment and pre-processes the collected data; The preprocessed data is fused and mapped into a multi-dimensional vector to generate a system state representation vector.

3. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 2, characterized in that: The target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption, and maximizing resource utilization, including: Minimize task response delay T delay The expression: in: is the set of all tasks, K is the total number of tasks, D k is the data volume of the kth task, f CPU Calculate the CPU frequency of the edge gateway, B link is the network link bandwidth, η net is the network congestion factor, Q queue is the task queue length, μ process is the task processing rate, ω k is the task priority weight, satisfying 4. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 3, characterized in that: The target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption and maximizing resource utilization, and also include: Minimize data processing cost C processing The expression: in: is the subset of tasks that need to be processed. For the subset of tasks that require data transmission, is the subset of tasks that need memory storage, c CPU 、c mem and c trans They are CPU computing power unit price, memory storage unit price and data transmission unit price respectively. and are the CPU time and memory capacity occupied by the kth task, D k is the data transmission amount of the kth task.

5. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 4, characterized in that: The target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption and maximizing resource utilization, and also include: Minimize energy consumption E consumption The expression: Where: T is the total number of running time windows, P CPU 、P net and P mem They are CPU computing power, network module power and memory power respectively. and are the CPU computing, data transmission and memory usage in time window t respectively.

6. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 5, characterized in that: The target parameters include minimizing task response delay, minimizing data processing cost, minimizing energy consumption and maximizing resource utilization, and also include: Maximize resource utilization U resource The expression: Where: γ CPU , γ mem and γ bw are the weight coefficients of CPU, memory and bandwidth respectively, T total 、S total and B total They are total CPU time, total memory and total bandwidth capacity, is the effective bandwidth utilization of the kth task.

7. The method for dynamically allocating computing power of an edge gateway for power equipment according to any one of claims 1 to 6, characterized in that: The constraints include computing resource usage not exceeding a preset upper limit, task response latency being lower than a real-time threshold, and energy consumption being lower than the device energy consumption limit, including: The computing power resource usage does not exceed the preset upper limit: Where: T max 、S max and B max These are the preset upper limits for CPU, memory, and bandwidth respectively; The task response latency is lower than the real-time threshold: Where: T threshold is the real-time threshold; Where: E limit The upper limit of device energy consumption.

8. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 1, characterized in that: The method uses the system state representation vector as input, solves the computing power dynamic allocation model through a reinforcement learning algorithm, and generates a task scheduling path and resource allocation ratio, including: The reinforcement learning model includes a state space, an action space and a reward function; wherein the state space is composed of the system state representation vector, including real-time computing power occupancy, task queue length, network bandwidth utilization, device energy consumption indicators and task priority weights; the action space includes discrete task scheduling path decisions and continuous resource allocation ratios.

9. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 8, characterized in that: The method uses the system state representation vector as input, solves the computing power dynamic allocation model through a reinforcement learning algorithm, and generates a task scheduling path and resource allocation ratio, and further includes: Input the current system state representation vector into a pre-trained deep reinforcement learning policy network, and output task scheduling path selection and resource allocation ratio decision; the task scheduling path is selected from a predefined set of candidate paths, including low-latency paths and high-throughput paths; the resource allocation ratio decision includes the allocation weights of CPU, memory, and bandwidth; If the selected scheduling path is a low-latency path, high-real-time tasks will be assigned to the low-latency path first, and sufficient network bandwidth resources will be reserved; if the selected scheduling path is a high-throughput path, routine monitoring tasks will be assigned to the high-throughput path, and computing and storage resources will be dynamically allocated according to the resource allocation weights; Based on the resource allocation weight, the CPU computing power, memory capacity and network bandwidth resources of the edge gateway are divided proportionally; Based on the actual task execution delay, resource utilization and energy consumption data, the reinforcement learning reward value is calculated, and the subsequent decision-making is optimized through the policy network parameter update mechanism.

10. The method for dynamically allocating computing power of an edge gateway of an electric power equipment according to claim 1, characterized in that: The step of optimizing the preset hierarchical buffer queue based on the task scheduling path and resource allocation ratio to generate a target dynamic allocation strategy includes: Based on the CPU and bandwidth weights in the resource allocation ratio, the scheduling frequency of high-real-time tasks in the priority queue is increased, where the high-real-time tasks are tasks whose response delay threshold is lower than a first preset value, and the scheduling weight of non-urgent tasks in the FIFO queue is reduced; if it is detected that the actual response delay of the high-real-time task reaches a second preset value, the capacity upper limit of the priority queue is increased, and some resources are preempted from the FIFO queue; The event-driven mechanism is triggered during the system idle period, and the backlog tasks in the FIFO queue are dynamically pulled for batch processing based on the currently available computing resources; The optimized queue scheduling rules and resource allocation ratio are determined as the target dynamic allocation strategy.

Images