New energy electric field resource scheduling method, system and device based on asynchronous task driven edge calculation and medium

Abstract

The invention discloses a new energy electric field resource scheduling method, system and device based on asynchronous task-driven edge computing and a medium, and the method comprises the steps: collecting system state data of a terminal device and an edge node, and constructing a task scheduling model; establishing a fusion environment; introducing a data timeliness index to construct a state updating model; establishing a resource scheduling optimization problem by taking optimization of system efficiency as a target; a deep reinforcement learning algorithm is adopted to convert into a sequence decision problem; and solving the sequence decision problem and outputting a resource scheduling strategy. According to the invention, by constructing a fusion environment of a task scheduling model and communication computing, the collaboration of computing resources among terminal equipment, edge nodes and adjacent terminal equipment is realized; information age indexes are introduced to quantify freshness, a resource scheduling optimization problem is converted into a sequence decision problem in combination with deep reinforcement learning, the system resource utilization rate, the data perception real-time performance and the autonomous decision-making ability in the environment are improved, and the operation efficiency and stability of the new energy electric field are effectively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of mobile edge computing technology, and in particular to a method, system, device and medium for scheduling new energy power plant resources based on asynchronous task-driven edge computing. Background Technology

[0002] With the rapid development of new energy industries such as wind power and photovoltaics, the scale of new energy power plants is constantly expanding, and the complexity of their operation management and resource scheduling is increasing. In order to ensure the efficient and stable operation of the power plant as a whole, it is necessary to perform real-time sensing, collaborative processing and resource allocation of the massive computing tasks generated by numerous heterogeneous terminal devices in the plant, such as wind turbines, photovoltaic inverters, and environmental sensors. Combining mobile edge computing technology to provide computing power near the data source has become an important technical direction for improving the operational efficiency of new energy power plants and reducing data transmission latency.

[0003] However, traditional scheduling schemes often employ static or centralized strategies, making it difficult to effectively coordinate heterogeneous computing resources among terminal devices, edge nodes, and neighboring terminal devices. When faced with task load fluctuations and terminal device mobility, the utilization efficiency of heterogeneous computing resources is low, and transmission bottlenecks are easily generated. In addition, existing new energy power plant resource scheduling methods often use traditional delay as an optimization indicator, which cannot accurately measure the overall timeliness of data from generation to processing completion. This results in insufficient real-time guarantee of state perception and control, affecting the timeliness and accuracy of power plant monitoring and control. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a new energy power plant resource scheduling method, system, device and medium based on asynchronous task-driven edge computing to solve the problems that existing scheduling schemes mostly adopt static or centralized strategies, which make it difficult to coordinate heterogeneous computing resources among terminal devices, edge nodes and neighboring terminal devices, resulting in low utilization, transmission bottlenecks, and reliance on traditional latency indicators, which cannot accurately measure the overall timeliness of data from generation to processing, thus affecting the real-time performance of perception and control.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a new energy power plant resource scheduling method based on asynchronous task-driven edge computing, comprising: Collect system status data from terminal devices and edge nodes, and construct a task scheduling model based on the system status data; Based on the task scheduling model, a fusion environment is established; In the fusion environment, a data timeliness index is introduced to construct a state update model; Combining the fusion environment and the state update model, a resource scheduling optimization problem is constructed with the goal of optimizing system performance; The resource scheduling optimization problem is transformed into a sequence decision problem using a deep reinforcement learning algorithm. By solving the sequence decision problem, a resource scheduling strategy is output.

[0007] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the step of constructing the task scheduling model includes: Collect task data and resource status of terminal devices, as well as computing resources of edge nodes, as system status data; The system status data is used to determine the task unloading ratio and the task unloading decision. A task scheduling model is constructed based on the task unloading ratio and task unloading decision.

[0008] The beneficial effects of this preferred technical solution are as follows: by determining the task offloading ratio and task offloading decision through system status data, the computing tasks are flexibly allocated and collaboratively processed among terminal devices, edge nodes and adjacent terminal devices, which improves the utilization rate of heterogeneous computing resources, reduces task processing latency and system energy consumption, and lays the foundation for subsequent resource scheduling optimization.

[0009] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the step of establishing a fusion environment according to the task scheduling model includes: The task scheduling model introduces direct communication technology between terminal devices and spectrum reuse technology. By employing the direct communication technology between terminal devices and the spectrum reuse technology, a converged environment is constructed that supports the offloading and collaborative processing of computing tasks between terminal devices and edge nodes, as well as between terminal devices themselves.

[0010] The beneficial effects of this preferred technical solution are as follows: by introducing direct communication and spectrum reuse technology between terminal devices, a decentralized, high-concurrency communication and computing converged environment is constructed, realizing flexible offloading and collaborative processing of computing tasks among terminal devices, edge nodes and adjacent terminal devices, reducing transmission latency and central load, and improving system spectrum efficiency and overall energy efficiency.

[0011] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the step of introducing a data timeliness index to construct a state update model includes: The task processing time data is obtained through the computing task processing process in the fusion environment; Based on the task processing time data, an information age index is introduced as a data timeliness indicator. Set a maximum tolerance delay constraint and construct a state update model based on the information age index.

[0012] The beneficial effects of this preferred technical solution are as follows: by introducing the information age index as a data timeliness index, the freshness of the calculation task from generation to completion can be quantified. Combined with the maximum tolerance delay constraint, a state update model is constructed, which overcomes the limitations of traditional delay indexes and improves the real-time performance and accuracy of state perception and control of new energy systems.

[0013] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the steps of constructing the resource scheduling optimization problem include: Combining the task offloading decision and the maximum tolerable delay constraint, a resource scheduling optimization problem is constructed with the goal of minimizing system latency and migration cost.

[0014] The beneficial effects of this preferred technical solution are as follows: by combining task offloading decision with maximum tolerable delay constraint, a resource scheduling optimization problem with the goal of minimizing system latency and migration cost is constructed, realizing global optimization under multiple constraints, balancing task processing efficiency and resource scheduling cost, and improving overall operational efficiency and economy.

[0015] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the step of transforming the resource scheduling optimization problem into a sequential decision problem includes: An asynchronous deep reinforcement learning algorithm is used to process the resource scheduling optimization problem, resulting in a sequence decision problem. Based on the sequence decision problem, determine the system state characteristics, scheduling control parameters, and performance evaluation function.

[0016] The beneficial effects of this preferred technical solution are as follows: by transforming the resource scheduling optimization problem into a sequential decision problem and using asynchronous deep reinforcement learning algorithms to handle high-dimensional environments, it can adaptively learn the optimal scheduling strategy under multi-variable coupling relationships, thereby improving autonomous decision-making ability and long-term operating efficiency.

[0017] As a preferred embodiment of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing described in this invention, the step of outputting the resource scheduling strategy includes: The system state characteristics, scheduling control parameters, and performance evaluation function of the sequential decision problem are used to train a deep reinforcement learning agent. Based on the deep reinforcement learning agent, a resource scheduling strategy is output.

[0018] The beneficial effects of this preferred technical solution are as follows: by training a deep reinforcement learning agent based on the three elements of a sequential decision problem, the resource scheduling strategy can be learned and optimized online autonomously, enabling the continuous output of resource scheduling strategies in complex and ever-changing new energy power field environments, thereby improving autonomy and environmental adaptability.

[0019] Secondly, the present invention provides a new energy power plant resource scheduling system based on asynchronous task-driven edge computing, comprising: The data acquisition module is used to collect system status data from terminal devices and edge nodes; The model building module is used to build task scheduling models, fusion environments, and state update models; The optimization processing module is used to construct the resource scheduling optimization problem and transform it into a sequence decision problem; The strategy execution module is used to output and execute resource scheduling strategies.

[0020] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing.

[0022] Compared with existing technologies, the beneficial effects of this invention are as follows: by constructing a fusion environment of task scheduling model and communication computing, the collaboration of computing resources among terminal devices, edge nodes and neighboring terminal devices is realized; by introducing the information age index to quantify freshness, and combining deep reinforcement learning to transform the resource scheduling optimization problem into a sequential decision problem, the system's resource utilization rate, data perception real-time performance and autonomous decision-making ability under the environment are improved, effectively ensuring the operating efficiency and stability of new energy power plants. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1This is a schematic diagram of the overall process of a new energy power plant resource scheduling method based on asynchronous task-driven edge computing according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the A3C algorithm architecture.

[0026] Figure 3 The graph shows the convergence performance of the A3C algorithm at different learning rates.

[0027] Figure 4 This chart compares the system costs of multiple schemes under different MEC server computing capabilities.

[0028] Figure 5 This is a comparison chart of system latency for different solutions under different terminal device scales. Detailed Implementation

[0029] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0030] Example 1, referring to Figures 1-5 As an embodiment of the present invention, a new energy power plant resource scheduling method based on asynchronous task-driven edge computing is provided, comprising: S100: Collect system status data of terminal devices and edge nodes, and construct a task scheduling model based on the system status data.

[0031] S200. Establish a fusion environment based on the task scheduling model.

[0032] S300. In the fusion environment, a data timeliness index is introduced to construct a state update model.

[0033] S400. Combining the fusion environment and the state update model, a resource scheduling optimization problem is constructed with the goal of optimizing system performance.

[0034] S500: The resource scheduling optimization problem is transformed into a sequence decision problem using a deep reinforcement learning algorithm.

[0035] S600. By solving the sequence decision problem, output the resource scheduling strategy.

[0036] It should be noted that the operating environment of new energy power plants is complex and variable. Traditional scheduling methods are difficult to cope with the fluctuation of task load and the heterogeneity of terminal equipment. Furthermore, relying on traditional delay indicators cannot accurately reflect the timeliness requirements of data processing, resulting in low resource utilization efficiency and insufficient real-time system response.

[0037] Therefore, to address the aforementioned challenges of resource coordination and insufficient data timeliness, the following steps (S100-S600) are employed: first, a task scheduling model is constructed to clarify the task offloading strategy; then, a converged communication and computing environment is established, and an information age index is introduced to quantify freshness; next, a resource scheduling optimization problem is constructed, and deep reinforcement learning is used to transform the resource scheduling optimization problem into a sequential decision problem and solve it, ultimately outputting a real-time and efficient resource scheduling strategy, providing complete technical support for the optimization and intelligent decision-making of new energy power plant resources.

[0038] Example 2, refer to Figures 1-5 As an embodiment of the present invention, based on the above embodiment, a new energy power plant resource scheduling method based on asynchronous task-driven edge computing is provided.

[0039] In this application embodiment, taking a large-scale wind farm as the application scenario, the specific implementation of the steps for constructing the task scheduling model for A1~A3 in S100 is as follows: A1. Collect task data and resource status of terminal devices, as well as computing resources of edge nodes, as system status data.

[0040] Specifically, assuming the wind farm contains 20 wind turbines (NEEs), each equipped with a vibration monitoring sensor and controller, the collected task data includes the size of the monitoring task. Required CPU cycles Maximum tolerable average information age (AoI) Task generation interval Resource status includes: NEE local computing power Remaining battery power Edge node computing resources Available bandwidth .

[0041] A2. Determine the task unloading ratio and task unloading decision based on the system status data.

[0042] Specifically, an energy-efficiency-optimized offloading strategy is adopted, defining a task offloading ratio coefficient. By solving the energy efficiency optimization problem: in, Total energy consumption, including transmission and computing energy consumption. The total latency includes both transmission and computation latency. This represents the task offloading ratio coefficient. The optimal offloading ratio is calculated based on the current channel state information and resource availability. This means that 70% of the workload needs to be uninstalled and processed. Task unloading decision Determined based on channel quality thresholds, when hour Unload to the MEC server, otherwise It is then processed locally in NEE and collaboratively processed via D2D link.

[0043] A3. Construct a task scheduling model based on the task unloading ratio and task unloading decision.

[0044] Specifically, the constructed task scheduling model is represented as a quadruple: in, A collection of terminal devices; For the set of task parameters; For the unloading ratio set; The task unloading decision set; the constraints of the task scheduling model include: .

[0045] In an optional implementation, step S100 may also employ a time-slot scheduling method, the steps of which are: based on the spatiotemporal distribution characteristics of task arrival, the scheduling period is divided into time slots of variable length, shorter time slots such as 0.5s are used during task-intensive periods to improve scheduling accuracy, and longer time slots such as 2s are used during task-sparse periods to reduce scheduling overhead.

[0046] In another optional implementation, step S100 may also employ a task unloading decision based on task priority, comprising the following steps: setting priority weights according to task type, and assigning weights to vibration monitoring tasks. Environmental data collection task weight Performance evaluation task weight When determining the unloading ratio, priority should be given to ensuring resource allocation for high-weight tasks.

[0047] In this embodiment of the application, step S200, the step of establishing the fusion environment according to the task scheduling model, includes B1~B2: B1. Through the task scheduling model, direct communication technology between terminal devices and spectrum reuse technology are introduced.

[0048] Specifically, based on the set of terminal devices in the task scheduling model Establish a terminal-to-terminal (D2D) link communication link. Divide the available 100MHz spectrum into 16 subcarriers using Orthogonal Frequency Division Multiplexing (OFDM), with each subcarrier having a bandwidth of... The channel power gain is calculated as follows: in, The small-scale fading component follows a complex Gaussian distribution with a mean of 0 and a variance of 1. The distance between terminal devices. This is the path loss factor. Indicates terminal device With terminal equipment Channel power gain between It is a small-scale fading component.

[0049] B2. By using the direct communication technology between terminal devices and the spectrum reuse technology, a fusion environment is constructed that supports the offloading and collaborative processing of computing tasks between terminal devices and edge nodes, as well as between terminal devices themselves.

[0050] Specifically, the constructed fusion environment includes three computing paths: The formula for calculating the transmission rate of the NEE to MEC server path is: Among them, the transmission power of terminal equipment Noise power spectral density , Terminal equipment The transfer rate to the MEC server, Terminal equipment The channel power gain to the MEC server is calculated in a similar way to that in B1. The calculated average transmission rate is 85Mbps. D2D Link Collaboration Path: When the task is unloaded, a decision is made. Furthermore, when nearby terminal devices are idle, the data is offloaded via the D2D link, and the transmission rate is calculated using the following formula: in, Terminal equipment To terminal equipment transmission rate, It refers to the channel power gain between terminal devices. For inter-cell interference, let's assume it is... The calculated average transmission rate is 45Mbps; Local compute path: The remaining tasks are processed locally in NEE, utilizing compute capabilities. .

[0051] In an optional implementation, step S200 may also employ an adaptive subcarrier allocation method, the steps of which are: allocating OFDM subcarriers according to channel state information, for better channel quality, i.e. The D2D link allocates more subcarriers, improving spectrum utilization.

[0052] In another optional implementation, step S200 may also include a distributed power control mechanism, comprising the steps of adjusting the transmit power according to the transmission distance for short-range D2D link communication. Use lower power This reduces system interference and energy consumption.

[0053] In this embodiment of the application, step S300, which involves introducing data timeliness indicators to construct a state update model, includes steps C1 to C3: C1. Obtain task processing time data through the computing task processing process in the fusion environment.

[0054] Specifically, based on the fusion environment established by S200, the task processing time of each computing path is monitored: Total latency of MEC server path: Total delay of D2D link path: Total local computation latency: .

[0055] C2. Using the task processing time data, introduce the information age index as a data timeliness indicator.

[0056] Specifically, define the terminal device In the time slot Information age AoI is: in, For terminal devices In the time slot Information age AoI, For time slots, The timestamp of the last successful data processing, generated according to the task interval. And process latency data to calculate the average information age (AoI): in, The average information age (AoI) is... Due to task processing delay, This represents the expectation operator.

[0057] C3. Set a maximum tolerance delay constraint and construct a state update model based on the information age index.

[0058] Specifically, set a maximum tolerance delay constraint. The constructed state update model is as follows: in, For each time slot The probability of generating data and the state update model satisfying the following constraints: .

[0059] In an optional implementation, step S300 may further employ an information age (AoI) weighting method based on task type, comprising the following steps: setting weights for information age (AoI) according to task criticality, and weighting vibration monitoring tasks accordingly. Environmental data collection task weight In the state update model, a weighted information age (AoI) is used. Optimize.

[0060] In another optional implementation, step S300 may also employ an adaptive generation probability adjustment mechanism, the steps of which are: adjusting the data generation probability according to the system load. When the average information age (AoI) is close to At that time, The efficiency was reduced from 0.6 to 0.4, which lowered the system load and ensured the timeliness of critical tasks.

[0061] In this embodiment of the application, step S400, the step of constructing the resource scheduling optimization problem, includes: Combining the task offloading decision and the maximum tolerable delay constraint, a resource scheduling optimization problem is constructed with the goal of minimizing system latency and migration cost.

[0062] Specifically, the resource scheduling optimization problem P is constructed as follows: Optimization goal: Constraints: in, For terminal devices The total task processing latency is 408.8ms based on the S300 MEC server path, 7111.1ms based on the D2D link path, and 3000ms based on the local computation latency. To mitigate migration costs, when service edge nodes change Otherwise, it is 0; This is a preference factor, with a value range of [0,1]. It is set according to the task type, such as vibration monitoring tasks. Environmental data collection ; The transmit power has a value range of [10, 23] dBm. For the allocated bandwidth resources, each subcarrier .

[0063] In an optional implementation, step S400 may also employ a weighted multi-objective optimization method, comprising the following steps: adjusting preference factors according to task urgency. When the system's average information age (AoI) exceeds 1.5 seconds, the preference factors for all tasks are... Increase by 0.2 to prioritize latency performance.

[0064] In another optional implementation, a distributed solution framework can also be used in step S400. The steps are as follows: the resource scheduling optimization problem is decomposed into multiple sub-problems, each edge node solves its local optimization problem independently, and then the global optimum is achieved through a coordination mechanism to obtain the resource scheduling optimization problem, thereby reducing the solution complexity.

[0065] In this embodiment of the application, step S500, which transforms the resource scheduling optimization problem into a sequential decision problem, includes steps E1 to E2: E1. The resource scheduling optimization problem is processed using an asynchronous deep reinforcement learning algorithm, resulting in a sequence decision problem.

[0066] Specifically, an asynchronous advantage action evaluation algorithm, such as the A3C algorithm, is adopted. The architecture diagram of the A3C algorithm is shown below. Figure 2 As shown, the resource scheduling optimization problem P is modeled as a Markov decision process (MDP), and its quintuple representation is as follows: in, For state space, For the action space, Let be the state transition probability. For the reward function, the discount factor .

[0067] E2. Based on the sequence decision problem, determine the system state characteristics, scheduling control parameters, and performance evaluation function.

[0068] Specifically, the system state characteristics are defined as follows: in, Let n be the channel gain from terminal device n to edge node k. Let n be the information age of the terminal device. The available computing resources for edge node k, The remaining energy of terminal device n is determined based on the remaining power level set in A1. calculate; The scheduling control parameters are defined as follows: in, The offloading ratio is adjusted based on channel conditions; Decisions for unloading real-time tasks; These are the transmit power control parameters; The performance evaluation function is defined as follows: in, As a penalty factor, This is an indicator function.

[0069] In an optional implementation, step S500 may also employ a hierarchical MDP modeling method, which involves dividing the decision-making process into two layers: the upper layer handles long-term resource reservation, and the lower layer handles real-time task scheduling, using state spaces and action spaces with different time scales respectively.

[0070] In another optional implementation, step S500 may also employ a course-based training strategy, which involves starting training from simplified scenarios such as fixed channels and uniform task distribution, and gradually increasing the environmental complexity to real wind farm scenarios to improve the convergence and stability of the algorithm.

[0071] In this embodiment of the application, step S600, the step of outputting the resource scheduling strategy, includes F1~F3: F1. Train a deep reinforcement learning agent using the system state characteristics, scheduling control parameters, and performance evaluation function of the sequential decision problem.

[0072] Specifically, an asynchronous advantage action evaluation algorithm, such as the A3C algorithm, is used to establish a training architecture with 16 parallel workers. Each worker maintains a local Actor-Critic network, and the parameters of the Actor-Critic network are as follows: Input layer dimension: 84 (corresponding to the system state feature dimension); Hidden layer: 2 fully connected layers, containing 256 and 128 neurons respectively; Output layer: The Actor-Critic network outputs the probability distribution of the scheduling control parameters, and the Critic network outputs the state value function estimate; The training hyperparameters are set as follows: learning rate Discount factor Entropy regularization coefficient The gradient clipping threshold is 0.5, and each worker collects 32 experience samples. Then, the policy gradient and value function gradient are calculated, and the Actor-Critic network parameters are updated asynchronously. like Figure 3 As shown, the convergence performance of the A3C algorithm in the edge computing system for task scheduling in new energy power plants is shown under different learning rates. When the learning rate is set to 0.0001 and 0.001, it exhibits good convergence performance as the training iterations proceed.

[0073] F2. Based on the deep reinforcement learning agent, output a resource scheduling strategy.

[0074] Specifically, the optimal policy network is trained. Receive real-time system status Output resource scheduling strategy: Resource scheduling policies include task offloading instructions, power control instructions, and resource reservation instructions, such as: For vibration monitoring device 1: unloading ratio Task unloading decision Offloaded to MEC server, transmit power ; For environmental monitoring equipment 5: unloading ratio Task unloading decision D2D link collaborative processing, transmit power ; Verification of the execution effect of resource scheduling strategy: like Figure 4 As shown, this paper compares the system costs of different schemes under different MEC server computing capabilities. Scheme 1 introduces task migration, D2D link collaboration, and AoI mechanisms. Scheme 2 is a no-migration scheme, where NEE tasks are scheduled to a MEC server but no task migration service is provided; the task must be completed on that MEC server. Scheme 3 is a no-migration, no-D2D link scheme, where NEE tasks are scheduled to a MEC server but no task migration service is provided, and inter-device communication is not considered. Scheme 4 is a task-only offloading scheme, where the edge server provides offloading services for MD within its communication coverage area, without considering the allocation of computing resources. Figure 4 It can be seen that Scheme 1 can fully leverage the heterogeneity and resource utilization of edge computing, maintaining the optimal system cost under different computing capabilities.

[0075] like Figure 5As shown, a comparison of system latency performance under different task sizes is presented. Average latency refers to the ratio of the total latency of unloading tasks to the number of tasks in the entire process. Figure 5 It can be seen that Scheme 1, by fully utilizing the multi-dimensional resources of the network, devices, and edge servers, achieves the best system latency performance, maintaining the lowest system latency performance under different task scales. Scheme 2, limited by the computing resources of a single MEC server, may have similar system latency performance to Scheme 1 for small-scale tasks, but as the task scale increases, local congestion leads to longer task waiting times, resulting in a significant increase in system latency performance. Scheme 3's system almost entirely relies on a single MEC server to process tasks; when the task scale increases, the system latency performance rises rapidly because there is no mitigation mechanism once the MEC server is overloaded. Scheme 4, lacking resource allocation optimization, is prone to bottlenecks when the task volume is large.

[0076] In an optional implementation, step S600 may also employ a policy optimization method based on transfer learning, which involves: using a policy network trained in other wind farm scenarios as the initial model, fine-tuning it with a small amount of real-time data from the current wind farm, and further training 1000 more models on the basis of the pre-trained model to improve the convergence speed.

[0077] In another optional implementation, step S600 may also employ an integrated multi-deep reinforcement learning agent collaboration mechanism, which involves: deploying multiple specialized deep reinforcement learning agents to handle different types of computational tasks, such as vibration monitoring tasks, environmental data acquisition tasks, and performance evaluation tasks; integrating the outputs of each deep reinforcement learning agent through a weighted voting mechanism to form the final resource scheduling strategy, thereby improving the robustness and accuracy of the resource scheduling strategy.

[0078] In summary, this invention constructs a new energy power plant resource scheduling system based on asynchronous task-driven edge computing, achieving intelligent collaborative and optimized management of heterogeneous computing resources in wind power scenarios. It collects system status data from terminal devices such as wind turbine monitoring sensors and edge nodes, establishes a task scheduling model, and determines the task offloading ratio and task offloading decisions. Then, it introduces D2D direct link communication and OFDM spectrum reuse technology to construct a fusion environment, supporting flexible task offloading between terminals, edges, and terminals. Furthermore, based on task processing time data, it introduces Information Age (AoI) as a data timeliness indicator, sets a maximum tolerable delay constraint, and establishes a state update model. Subsequently, it combines task offloading decisions and the maximum tolerable delay constraint to construct a resource scheduling optimization problem aimed at minimizing system latency and migration costs. Using asynchronous deep reinforcement learning algorithms such as A3C, it transforms the resource scheduling optimization problem into a sequential decision problem, defining the system state, action space, and reward function. Finally, it trains a deep reinforcement learning agent to output real-time resource scheduling strategies, achieving online optimization and autonomous decision-making in resource allocation.

[0079] Example 3 illustrates a schematic scheme for a new energy power plant resource scheduling method based on asynchronous task-driven edge computing. It should be noted that the technical solution of this new energy power plant resource scheduling system based on asynchronous task-driven edge computing belongs to the same concept as the technical solution of the aforementioned new energy power plant resource scheduling method based on asynchronous task-driven edge computing. Details not described in detail in this embodiment can be found in the description of the aforementioned new energy power plant resource scheduling method based on asynchronous task-driven edge computing.

[0080] This embodiment also provides a new energy power plant resource scheduling system based on asynchronous task-driven edge computing, including: The data acquisition module is used to collect system status data from terminal devices and edge nodes; The model building module is used to build task scheduling models, fusion environments, and state update models; The optimization processing module is used to construct the resource scheduling optimization problem and transform it into a sequence decision problem; The strategy execution module is used to output and execute resource scheduling strategies.

[0081] This embodiment also provides an electronic device applicable to the scheduling of new energy power plant resources based on asynchronous task-driven edge computing, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the new energy power plant resource scheduling method based on asynchronous task-driven edge computing proposed in the above embodiment.

[0082] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the new energy power plant resource scheduling method based on asynchronous task-driven edge computing proposed in the above embodiment.

[0083] The storage medium proposed in this embodiment and the new energy power plant resource scheduling method based on asynchronous task-driven edge computing proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0084] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0085] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A new energy power plant resource scheduling method based on asynchronous task-driven edge computing, characterized in that, include: Collect system status data from terminal devices and edge nodes, and construct a task scheduling model based on the system status data; Based on the task scheduling model, a fusion environment is established; In the fusion environment, a data timeliness index is introduced to construct a state update model; Combining the fusion environment and the state update model, a resource scheduling optimization problem is constructed with the goal of optimizing system performance; The resource scheduling optimization problem is transformed into a sequence decision problem using a deep reinforcement learning algorithm. By solving the sequence decision problem, a resource scheduling strategy is output.

2. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 1, characterized in that, The steps to build a task scheduling model include: Collect task data and resource status of terminal devices, as well as computing resources of edge nodes, as system status data; The system status data is used to determine the task unloading ratio and the task unloading decision. A task scheduling model is constructed based on the task unloading ratio and task unloading decision.

3. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 2, characterized in that, Based on the task scheduling model, the steps for establishing a fusion environment include: The task scheduling model introduces direct communication technology between terminal devices and spectrum reuse technology. By employing the direct communication technology between terminal devices and the spectrum reuse technology, a converged environment is constructed that supports the offloading and collaborative processing of computing tasks between terminal devices and edge nodes, as well as between terminal devices themselves.

4. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 3, characterized in that, The steps for constructing a state update model by introducing data timeliness indicators include: The task processing time data is obtained through the computing task processing process in the fusion environment; Based on the task processing time data, an information age index is introduced as a data timeliness indicator. Set a maximum tolerance delay constraint and construct a state update model based on the information age index.

5. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 4, characterized in that, The steps involved in constructing a resource scheduling optimization problem include: Combining the task offloading decision and the maximum tolerable delay constraint, a resource scheduling optimization problem is constructed with the goal of minimizing system latency and migration cost.

6. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 5, characterized in that, The steps to transform the resource scheduling optimization problem into a sequential decision problem include: An asynchronous deep reinforcement learning algorithm is used to process the resource scheduling optimization problem, resulting in a sequence decision problem. Based on the sequence decision problem, determine the system state characteristics, scheduling control parameters, and performance evaluation function.

7. The new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in claim 6, characterized in that, The steps for outputting the resource scheduling policy include: The system state characteristics, scheduling control parameters, and performance evaluation function of the sequential decision problem are used to train a deep reinforcement learning agent. Based on the deep reinforcement learning agent, a resource scheduling strategy is output.

8. A new energy power plant resource scheduling system based on asynchronous task-driven edge computing, employing the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect system status data from terminal devices and edge nodes; The model building module is used to build task scheduling models, fusion environments, and state update models; The optimization processing module is used to construct a resource scheduling optimization problem and transform it into a sequence decision problem. The strategy execution module is used to output and execute resource scheduling strategies.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the new energy power plant resource scheduling method based on asynchronous task-driven edge computing as described in any one of claims 1 to 7.

Images