Method for cross-core task scheduling and energy efficiency coordination based on multi-core controller

Abstract

This application discloses a cross-core task scheduling and energy efficiency coordination method based on a multi-core controller, belonging to the field of wind power generation control technology. The method includes: triggering a dynamic computing power reconfiguration mechanism when current conditions meet the criteria of processing core degradation or resource constraints; automatically readjusting the graph structure of the task dependency graph based on the dynamic computing power reconfiguration mechanism and performing real-time repair on the task scheduling scheme; and feeding back the task completion status, real-time indicators, and energy consumption data during the execution phase to the reinforcement learning engine and heterogeneous perception module to update the model parameters of the task scheduling strategy and the task dependency model, resulting in an updated task scheduling strategy and an updated task dependency model. This processing method, by introducing parallel execution on different processing cores of the multi-core controller, can significantly improve processing efficiency.

Description

Technical Field

[0001] This application belongs to the field of wind power generation control technology, and specifically relates to a processing method for cross-core task scheduling and energy efficiency coordination based on a multi-core controller. Background Technology

[0002] The main control system of a wind turbine is its brain, responsible for collecting external environmental data such as wind speed and direction, as well as internal status information. It calculates optimal operating commands through preset control algorithms to achieve maximum energy capture and ensure equipment safety. The wind turbine main control system needs to execute multiple critical tasks in parallel, including: nacelle status monitoring, sensor fault detection, yaw control, pitch control, start / stop control, grid connection / disconnection control, data acquisition and statistics, and remote communication. With the increase in single-unit capacity and control complexity of wind turbines, the software functions of the main control system are becoming increasingly rich (e.g., adding vibration monitoring, load optimization, predictive maintenance algorithms), significantly increasing the number of real-time tasks and the amount of data that need to be processed.

[0003] Traditional wind power main control controllers mostly use a single CPU or a simple PLC controller to schedule tasks serially or on a timer, which has the following limitations: Computing bottlenecks and inefficient utilization: Single-core processing struggles to handle the ever-increasing demands of parallel computing, easily leading to CPU overload or requiring increased clock speeds to meet requirements, resulting in additional energy consumption and thermal stress. Even when some mainframe hardware has multiple cores, the lack of effective task allocation strategies often results in some cores being busy while others are idle, failing to fully utilize computing resources.

[0004] Fixed scheduling strategies: Traditional master control software often uses fixed-cycle or fixed-priority scheduling logic, lacking the ability to adjust according to real-time load changes. When encountering abnormal operating conditions (such as: drastic wind speed fluctuations, fault conditions), fixed scheduling strategies may not be able to respond in time to changes in task execution time and priority, resulting in delays in critical control tasks or even missing deadlines.

[0005] The Challenges of Multi-Core Heterogeneous Architecture: Next-generation embedded control chips often adopt heterogeneous multi-core architectures, integrating cores with varying performance (e.g., high-performance cores + low-power cores) and even dedicated acceleration units. While heterogeneous architectures increase the upper limit of system performance, they also introduce complexity in task mapping and resource coordination.

[0006] Insufficient energy efficiency management: Long-term operation of wind turbines also places demands on the power consumption and heat dissipation of the main control system. Traditional dispatching does not consider energy consumption optimization, often resulting in all core components running continuously at full speed, causing component overheating and reducing system reliability.

[0007] Reliable Operation and Fault Tolerance: Wind turbine main control systems have extremely high reliability requirements; any abnormal execution of control tasks can affect power generation or cause a safe shutdown. Traditional solutions mainly rely on hardware redundancy and simple monitoring alarms. If any part of the main control fails, it may lead to a complete turbine shutdown. In a multi-core environment, if a core or task fails, there is a lack of mechanisms to quickly switch the task to other cores to continue running. Fixed scheduling also struggles to respond promptly to abnormal situations such as abnormal sensor data and network communication delays.

[0008] In summary, existing scheduling methods for wind turbine main controllers are insufficient to fully utilize multi-core computing power and cannot balance real-time performance, energy efficiency, and reliability in dynamic environments. This necessitates a novel multi-core scheduling method for wind turbine main controllers that can dynamically allocate tasks based on wind turbine operating characteristics, while simultaneously optimizing energy use and improving system fault tolerance, thus overcoming the limitations of traditional technologies. Summary of the Invention

[0009] To address the aforementioned issues, this application provides a processing method, system, storage medium, and electronic device based on a multi-core controller for cross-core task scheduling and energy efficiency coordination, which can significantly improve processing efficiency.

[0010] In a first aspect, embodiments of this application provide a cross-core task scheduling and energy efficiency coordination method based on a multi-core controller, the method comprising: The heterogeneous sensing module collects multi-source input parameters of the wind power main control system in real time, including environmental parameters and operating status parameters. A task dependency graph for wind power main control tasks is constructed based on a graph structure modeling method, and a task dependency model is generated by taking each control task as a node and the dependency relationship between tasks as an edge. In the task dependency graph, each control task is taken as a node and the dependency relationship between tasks is taken as an edge. The state space and action space of reinforcement learning are constructed based on the dependency graph, and a reward and punishment mechanism including a multi-objective optimization function is configured. The optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication latency, and stabilizing the energy consumption function curve. According to the task scheduling strategy corresponding to the reinforcement learning engine, each control task in the task dependency graph is assigned to different processing cores of the multi-core controller for parallel execution, thereby obtaining the task completion status of the execution phase. When the current conditions are detected to meet the conditions of processing core degradation or resource limitation, the computing power dynamic reconstruction mechanism is triggered, and the graph structure of the task dependency graph is automatically readjusted based on the computing power dynamic reconstruction mechanism and the task scheduling scheme is repaired in real time. The task completion status, real-time indicators, and energy consumption data during the execution phase are fed back to the reinforcement learning engine and the heterogeneous perception module to obtain feedback results. Based on the feedback results, the model parameters of the task scheduling strategy and the task dependency model are updated to obtain the updated task scheduling strategy and the updated task dependency model.

[0011] Optionally, after acquiring the multi-source input parameters of the wind power main control system in real time through the heterogeneous sensing module, the method further includes: The multi-source input parameters are obtained, and the environmental parameters include at least: wind speed, wind direction, ambient temperature, and power grid load; the operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and the utilization rate and temperature of each processing core of the main control multi-core controller. Obtain wind conditions and the operating status of the wind power main control system; The update cycle of each parameter in the multi-source input parameters is dynamically adjusted based on the wind conditions and the system operating status.

[0012] Optionally, after constructing the task dependency graph of the wind power master control task using the graph structure modeling method, the method further includes: Obtain the task dependency graph, which is constructed based on a directed acyclic graph model. In the task dependency graph, each node represents a wind power master control task and multiple attribute information corresponding to the wind power master control task. The multiple attribute information includes: the execution cycle, deadline, and expected execution time of the task. The directed edge between any two nodes in each node represents the data relationship or control dependency relationship between the tasks. Obtain the node attribute information of each node in the task dependency graph; The task dependency graph and the node attribute information of each node are input into the reinforcement learning engine for processing to obtain the processing result.

[0013] Optionally, the step of constructing the state space and action space of reinforcement learning based on the dependency graph, and configuring a reward and punishment mechanism including a multi-objective optimization function, includes: The task scheduling process is transformed into a Markov decision process through the reinforcement learning engine. The state space and action space of reinforcement learning are constructed based on the dependency graph. The state space includes: the queue of ready tasks in the current task dependency graph, the running load status of each processing core, and the remaining time of the task deadline. The action space includes: scheduling decisions, which are scheduling decisions used to select target processing cores for ready tasks or adjust the execution priority of tasks. The configuration includes a reward and penalty mechanism for multi-objective optimization functions; Obtain the reward and punishment function used by the reward and punishment mechanism; Based on the influence of multiple indicators, the reward and penalty function provides a positive reward for the first scheduling behavior that meets the first condition and imposes a penalty for the second scheduling behavior that meets the second condition, so as to guide the scheduling strategy to converge in the direction of satisfying multi-objective optimization. The multiple indicators include: real-time failure, communication latency, and energy consumption fluctuation; the first condition includes: completing the task on time, or reducing inter-core communication latency, or stabilizing the energy consumption curve; the second condition includes: completing the task beyond the deadline, or introducing high communication latency, or drastic energy consumption fluctuation.

[0014] Optionally, after the configuration includes a reward and penalty mechanism for a multi-objective optimization function, the method further includes: Obtain the operating status of the wind power main control system and the priority control targets; Obtain the sub-objectives, which include: real-time performance, communication latency, and stable energy consumption; Based on the operating conditions and the priority control objectives, the weights of each sub-objective are dynamically adjusted.

[0015] Optionally, dynamically adjusting the weights of each sub-objective based on the operating conditions and the priority control objectives includes: If a deterioration in a real-time performance indicator is detected, the weight of that real-time performance indicator will be increased. If an increase in communication latency is detected, the weight corresponding to the communication latency indicator will be increased. If an increase in energy consumption fluctuation is detected, the weight of the energy consumption stability index will be increased.

[0016] Secondly, embodiments of this application provide a processing system for cross-core task scheduling and energy efficiency coordination based on a multi-core controller, the system comprising: The heterogeneous sensing module is used to collect multi-source input parameters of the wind power main control system in real time. The multi-source input parameters include environmental parameters and operating status parameters. The first building module is used to construct the task dependency graph of the wind power main control task based on the graph structure modeling method; The generation module is used to generate a task dependency model by taking each control task as a node and the dependencies between tasks as edges. In the task dependency graph, each control task is taken as a node and the dependencies between tasks are taken as edges. The second construction module is used to construct the state space and action space of reinforcement learning based on the dependency graph. The configuration module is used to configure a reward and punishment mechanism including a multi-objective optimization function, wherein the optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication delay, and stabilizing the energy consumption function curve; The parallel execution module is used to allocate each control task in the task dependency graph to different processing cores of the multi-core controller for parallel execution according to the task scheduling strategy corresponding to the reinforcement learning engine, so as to obtain the task completion status of the execution phase. The triggering module is used to trigger the dynamic reconfiguration mechanism of computing power when the current conditions are detected to meet the conditions of core degradation or resource limitation. The adjustment and repair module is used to automatically readjust the graph structure of the task dependency graph based on the computing power dynamic reconstruction mechanism and to repair the task scheduling scheme in real time. The feedback module is used to feed back the task completion status, real-time indicators and energy consumption data of the execution phase to the reinforcement learning engine and the heterogeneous perception module to obtain feedback results. Based on the feedback results, the model parameters of the task scheduling strategy and the task dependency model are updated to obtain the updated task scheduling strategy and the updated task dependency model.

[0017] Optionally, the system further includes: The first acquisition module is used to acquire the multi-source input parameters after real-time acquisition of the multi-source input parameters of the wind power main control system. The environmental parameters include at least: wind speed, wind direction, ambient temperature, and grid load. The operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and utilization rate and temperature of each processing core of the main control multi-core controller. The second acquisition module is used to acquire information based on wind conditions and system operating status. The dynamic adjustment module is used to dynamically adjust the update cycle of each parameter in the multi-source input parameters according to the wind conditions and the system operating status.

[0018] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.

[0019] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.

[0020] Compared with the prior art, this application has the following advantages: It can greatly improve processing efficiency.

[0021] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart is shown for a cross-core task scheduling and energy efficiency coordination processing method based on a multi-core controller according to an embodiment of this application; Figure 2 A schematic diagram of the structure of a processing system 200 based on a multi-core controller for cross-core task scheduling and energy efficiency coordination according to an embodiment of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] This application provides a processing method and system for cross-core task scheduling and energy efficiency coordination based on a multi-core controller, as well as a computer-readable medium and an electronic device, which will be described below with reference to the accompanying drawings.

[0026] Please refer to Figure 1 It illustrates flowcharts of cross-core task scheduling and energy efficiency coordination processing methods based on multi-core controllers provided in some embodiments of this application, such as... Figure 1 As shown, the multi-core controller-based cross-core task scheduling and energy efficiency coordination processing method may include the following steps: Step S101: Collect multi-source input parameters of the wind power main control system in real time through the heterogeneous sensing module. The multi-source input parameters include environmental parameters and operating status parameters.

[0027] In one example, after acquiring multi-source input parameters of the wind power main control system in real time through the heterogeneous sensing module, the processing method provided in this application embodiment further includes the following steps: Acquire multi-source input parameters. Environmental parameters include at least: wind speed, wind direction, ambient temperature, and grid load. Operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and utilization rate and temperature of each processing core of the main control multi-core controller. Obtain wind conditions and the operating status of the wind power main control system; The update cycle of each parameter in the multi-source input parameters is dynamically adjusted according to wind conditions and system operating status.

[0028] In the processing method provided in this application embodiment, the parameter update cycle of the heterogeneous sensing module is dynamically adjusted according to the wind conditions and system operating status. Specifically, the update cycles of the environmental parameters and operating status parameters are set respectively and adaptively adjusted according to their rate of change to ensure the real-time performance and accuracy of the sensing data. The collected environmental parameters and operating status parameters are combined to form sensing data and output to the reinforcement learning engine for subsequent scheduling decisions.

[0029] Step S102: Construct a task dependency graph of the wind power main control task based on the graph structure modeling method, and generate a task dependency model by taking each control task as a node and the dependency relationship between tasks as an edge. In the task dependency graph, each control task is taken as a node and the dependency relationship between tasks is taken as an edge.

[0030] In the processing method provided in the embodiments of this application, the wind power main control task includes the control tasks of wind turbine subsystems such as blade pitch control, nacelle yaw control, generator power regulation, converter control and braking system control. Each control task establishes a corresponding dependency connection in the task dependency relationship graph based on its functional logic and data interaction relationship.

[0031] In one example, after constructing the task dependency graph of the wind power master control task based on the graph structure modeling method, the processing method provided in this application embodiment further includes the following steps: Obtain the task dependency graph, which is constructed based on a directed acyclic graph model. In the task dependency graph, each node represents a wind power master control task and multiple attribute information corresponding to that wind power master control task. The multiple attribute information includes: the execution cycle, deadline, and expected execution time of the task; the directed edge between any two nodes in each node represents the data relationship or control dependency relationship between the tasks. Obtain the node attribute information of each node in the task dependency graph; The task dependency graph and the node attribute information of each node are input into the reinforcement learning engine for processing to obtain the processing results.

[0032] In the processing method provided in this application embodiment, the task dependency graph is constructed using a directed acyclic graph model. Each node represents a wind power master control task and includes attribute information such as the task's execution cycle, deadline, and expected execution time. The directed edges between nodes represent the data or control dependencies between tasks. The scheduling input interface inputs the constructed task dependency graph and its node attribute information into the reinforcement learning scheduling engine as the basis for task scheduling decisions.

[0033] Step S103: Construct the state space and action space of reinforcement learning based on the dependency graph, and configure a reward and punishment mechanism including a multi-objective optimization function. The optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication latency, and stabilizing the energy consumption function curve.

[0034] In one example, the state and action spaces for reinforcement learning are constructed based on a dependency graph, and a reward and punishment mechanism including a multi-objective optimization function is configured, including the following steps: By using a reinforcement learning engine, the task scheduling process is transformed into a Markov decision process; The state space and action space of reinforcement learning are constructed based on the dependency graph. The state space includes: the queue of ready tasks in the current task dependency graph, the running load status of each processing core, and the remaining time before the task deadline. The action space includes: scheduling decisions, which are scheduling decisions used to select target processing cores for ready tasks or adjust the execution priority of tasks. The configuration includes a reward and penalty mechanism for multi-objective optimization functions; Obtain the reward and punishment functions used by the reward and punishment mechanism; Based on multiple indicators, a reward and penalty function is used to give positive rewards to the first scheduling behavior that meets the first condition and to impose penalties on the second scheduling behavior that meets the second condition, so as to guide the scheduling strategy to converge in the direction of satisfying multi-objective optimization. The multiple indicators include: real-time failure, communication latency, and energy consumption fluctuation; the first condition includes: completing the task on time, or reducing inter-core communication latency, or stabilizing the energy consumption curve; the second condition includes: completing the task beyond the deadline, or introducing high communication latency, or drastic energy consumption fluctuation.

[0035] In the processing method provided in this application embodiment, the reinforcement learning engine models the task scheduling process as a Markov decision process. Its state space includes information such as the ready task queue in the current task dependency graph, the running load status of each processing core, and the remaining time before the task deadline. The action space includes scheduling decisions such as selecting a target processing core for ready tasks or adjusting the execution priority of tasks. The reward and penalty function of the reinforcement learning engine comprehensively considers indicators such as real-time failure, communication latency, and energy consumption fluctuations. It gives positive rewards to scheduling behaviors that complete tasks on time, reduce inter-core communication latency, and stabilize the energy consumption curve, and imposes penalties on scheduling behaviors that exceed the deadline or introduce high communication latency and drastic energy consumption fluctuations, thereby guiding the scheduling strategy to converge in the direction of satisfying multi-objective optimization.

[0036] In one example, after configuring a reward and penalty mechanism that includes a multi-objective optimization function, the processing method provided in this application embodiment further includes the following steps: Obtain the operating status of the wind power main control system and the priority control targets; Obtain the sub-objectives, which include: real-time performance, communication latency, and stable energy consumption; The weights of each sub-objective are dynamically adjusted based on operating conditions and priority control objectives.

[0037] In one example, the weights of each sub-objective are dynamically adjusted based on operating conditions and priority control objectives, including the following steps: If a deterioration in a real-time performance indicator is detected, the weight of that real-time performance indicator will be increased. If an increase in communication latency is detected, the weight corresponding to the communication latency indicator will be increased. If an increase in energy consumption fluctuation is detected, the weight of the energy consumption stability index will be increased.

[0038] In the processing method provided in this application embodiment, the multi-objective optimization function adopts a weight coordination adjustment strategy for each sub-objective of real-time performance, communication delay, and stable energy consumption. That is, the weight coefficients of each sub-objective are dynamically adjusted according to the operating conditions of the wind power main control system and the priority control objectives. When a deterioration in real-time performance is detected, the weight of the real-time performance objective is increased; when communication delay increases, the weight of the communication delay objective is increased; and when the energy consumption fluctuation amplitude increases, the weight of the stable energy consumption objective is increased. This balances the conflicts between the optimization objectives under different operating conditions and ensures the comprehensive performance of scheduling optimization.

[0039] Step S104: According to the task scheduling strategy corresponding to the reinforcement learning engine, the control tasks in the task dependency graph are allocated to different processing cores of the multi-core controller for parallel execution, and the task completion status of the execution phase is obtained.

[0040] Step S105: When the current conditions are detected to meet the conditions of processing core degradation or resource limitation, the computing power dynamic reconstruction mechanism is triggered, and the graph structure of the task dependency graph is automatically readjusted based on the computing power dynamic reconstruction mechanism, and the task scheduling scheme is repaired in real time.

[0041] In the processing method provided in the embodiments of this application, an inter-core communication path planning algorithm is used to optimize communication latency for cross-processing core task data transmission: tasks with direct data dependencies are preferentially allocated to processing cores with high-speed interconnect channels to shorten the inter-core communication distance and time; for tasks that must be executed on different processing cores and need to communicate with each other, the shortest communication path is planned and the communication timing is scheduled in a time-sharing manner to avoid bus contention conflicts, thereby reducing communication latency and bandwidth overhead.

[0042] In the processing method provided in this application embodiment, an energy consumption function curve smoother is set to perform real-time smoothing of the power consumption curve of the multi-core controller. The smoother monitors the instantaneous power consumption of each processing core and predicts the load change trend, dynamically adjusts the task scheduling rhythm or the working frequency of the processing core, and avoids the energy consumption spikes caused by the simultaneous execution of high-power tasks. By fine-tuning and peak-shifting the task execution sequence, the overall energy consumption curve over time becomes smoother, reducing the impact of energy consumption fluctuations on system stability.

[0043] In the processing method provided in this application embodiment, a fault adaptive reconfiguration scheduling mechanism is set up. When a processing core is detected to have experienced performance degradation or failure, or when the system resource utilization rate exceeds a preset threshold, the reconfiguration scheduling mechanism is triggered. During the reconfiguration process, affected processing cores are removed or downgraded in real time, and the task dependency graph is adjusted. Tasks originally allocated to the affected cores are remapped and allocated to other normally functioning cores, and an updated task scheduling scheme is dynamically generated. During the scheduling recovery process, critical control tasks are prioritized for migration and continued execution to ensure the continuity and stability of the wind power main control system control process.

[0044] In the processing method provided in this application embodiment, a special scheduling weight adjustment strategy is implemented when the wind farm control subsystem is under high load conditions. High load conditions include situations where extreme winds cause a significant increase in the load of pitch control and yaw control tasks. In this case, the weight or priority of key tasks such as pitch control and yaw control in scheduling optimization is increased to ensure that these key control tasks obtain core processing resources with higher priority and shorten response time. Even if the weight of other objectives (such as energy consumption optimization) is temporarily reduced as a result, the safety and stability of wind turbine operation are given priority.

[0045] Step S106: Feed back the task completion status, real-time indicators and energy consumption data of the execution phase to the reinforcement learning engine and heterogeneous perception module to obtain feedback results. Based on the feedback results, update the model parameters of the task scheduling strategy and the task dependency model to obtain the updated task scheduling strategy and the updated task dependency model.

[0046] The processing method provided in this application forms a closed loop of the entire process, including perception, modeling, decision-making, scheduling, execution, and feedback.

[0047] The processing method provided in this application, based on a reinforcement learning-based scheduling engine, comprehensively considers three aspects: real-time performance, communication latency, and stable energy consumption. It can autonomously optimize scheduling schemes in complex multi-core environments, significantly improving system real-time performance and energy efficiency compared to traditional fixed strategies. For example, on a real-time control platform, reinforcement learning scheduling increased the on-time task completion rate from 92% to 99.6%, while reducing power consumption by approximately 22%; thus achieving near-zero default rate real-time assurance and significant energy savings.

[0048] The processing method provided in this application uses task dependency graph modeling to enable parallel scheduling and maximize the utilization of multi-core parallel capabilities. At the same time, the introduction of the directed acyclic graph model endows the system with fault tolerance and self-healing capabilities. When some tasks or cores fail, there is no need for a global shutdown and restart. Normal operation can be restored simply by locally rescheduling the affected task nodes, ensuring the continuous and stable operation of the wind turbine control.

[0049] The processing method provided in this application addresses the challenges of processor core performance fluctuations or hardware degradation by offering a dynamic reconfiguration mechanism. This mechanism detects resource changes in real time and adaptively adjusts the computing load allocation, preventing system failures caused by single-point performance bottlenecks or faults. Compared to pre-static allocation strategies, the processing method provided in this application can rapidly reorganize computing power in unexpected situations, ensuring that control tasks are executed as required, significantly improving the reliability and fault tolerance of the wind turbine main control system.

[0050] The processing method provided in this application constructs a closed-loop scheduling system with feedback perception throughout the entire process, overcoming the limitations of traditional open-loop scheduling in adapting to changes in a timely manner. Through closed-loop feedback, the scheduling algorithm can continuously learn from environmental changes and historical performance, constantly adjusting and optimizing the scheduling strategy, enabling the system performance to improve autonomously over time, exhibiting self-learning and self-optimization characteristics. This online evolution capability of the scheduling strategy is a novel approach in the field of wind power main control, producing unexpected technical effects.

[0051] The processing method provided in this application, while introducing a reinforcement learning scheduling engine, retains and strengthens supporting modules such as energy management and communication optimization, which operate in conjunction with the scheduling engine. The energy management module controls the power consumption of each core through dynamic voltage and frequency adjustment, ensuring the stability of the energy consumption function curve. The communication optimization module optimizes cross-core data transmission based on task dependencies (e.g., scheduling high-frequency interactive tasks to the same or adjacent cores to reduce communication latency). These modules are organically integrated with RL scheduling decisions: on the one hand, they provide timely feedback on energy consumption and communication overhead for reinforcement learning; on the other hand, they execute corresponding optimization actions in real time based on Markov decisions, thereby achieving a synergistic improvement in performance and energy efficiency across the overall architecture. The processing method provided in this application has good flexibility and scalability, and can smoothly expand the task graph model and optimize the target weight configuration according to different wind farm scales and control requirements, adapting to constantly changing operational requirements.

[0052] In summary, the processing method provided in this application combines task graph modeling, reinforcement learning intelligent decision-making, multi-objective optimization, and dynamic self-healing scheduling, which significantly improves the real-time performance, reliability, and energy utilization efficiency of the wind turbine main control system in a multi-core heterogeneous environment, and has outstanding innovation and practical value.

[0053] Example The cross-core task scheduling and energy efficiency coordination method based on a multi-core controller provided in this application includes the following steps: The operating status parameters and environmental parameters of the wind power main control system are acquired in real time through the heterogeneous sensing module. Based on the graph structure modeling method, a dependency graph of wind power main control tasks is constructed, with each control task as a node and the dependency relationship between tasks as an edge, to generate a task dependency model; In the above scheme, firstly, the control task list and its attributes (cycle, deadline, estimated execution time, priority, task category and input / output signals, etc.) of the wind power master control are summarized. Each control task is abstracted into a node. Directed edges are generated based on data / causal relationships (such as the output of the upstream task serving as the input of the downstream task) and necessary temporal sequence relationships to form an acyclic task dependency graph. Then, feature vectors (such as cycle, deadline, execution time, priority, category encoding, etc.) are constructed for the nodes and the edges are weighted (such as cross-core communication latency, data volume / bandwidth requirements, temporal urgency, etc.) to obtain the adjacency matrix (or weighted adjacency matrix) and topological order. Finally, a task dependency model containing a node set, edge set, node feature matrix and edge weight matrix is ​​produced, which is used as a unified input for the subsequent reinforcement learning engine and scheduling decision engine, and supports dynamic reconstruction and rapid updating when the core degrades / task changes.

[0054] The search for task scheduling strategies is driven by a reinforcement learning engine. The state space and action space of reinforcement learning are constructed based on the task dependency graph. A reward and punishment mechanism with a multi-objective optimization function is designed. The multi-objective optimization function aims to minimize the real-time default rate, minimize the inter-core communication latency, and smooth the energy consumption function curve.

[0055] In the above scheme, the state space of the reinforcement learning engine includes: a graph structure embedding representation of the task dependency graph; real-time task parameters (deadline, remaining execution time); the current utilization and temperature status of each core of the multi-core controller; the operating environment parameters of the wind power main control system (wind speed, wind direction, grid load); and power consumption prediction values.

[0056] The action space of the reinforcement learning engine includes: scheduling decisions based on task-core pairs, and combining the operating frequency level of the core to form joint actions; wherein, the frequency level is a predefined set of discrete levels used to control the operating frequency and voltage of the core.

[0057] The design of the reward and punishment function of the reinforcement learning engine includes three sub-objectives: real-time default rate, inter-core communication latency, and the smoothness of the energy consumption function curve. The default detection of deadline, the accumulation of communication path overhead, and the variance evaluation of the power consumption curve sliding window are used for modeling, respectively. The three are then synthesized into a scheduling reward signal through normalization and dynamic weighting.

[0058] reward function Using weighted sum form: Among them, real-time default rate reward , This represents the number of tasks within the sliding window that have exceeded their deadline. Total number of tasks within this window; Inter-core communication latency penalty , For task-dependent edge sets, For the task With the task Communication path delay between them The normalization constant is Indicates task With the task They are scheduled to different processing cores, otherwise they are... Energy consumption function curve stability penalty ,in, , For time points System power consumption, For the length of the sliding window, The window starts at time 1; the weights satisfy the following conditions: and The dynamic weight adjustment logic is as follows: when season , when season And proportionally reduce the remaining weights to maintain ,in, and For preset threshold, The step size parameter is used; all the above statistics are within a sliding window. Internal calculation.

[0059] Based on the task scheduling strategy obtained by the reinforcement learning engine, each control task in the task dependency graph is assigned to different processing cores of the multi-core controller for parallel execution. Communication paths between tasks across processing cores are planned to reduce communication latency, and the execution sequence of tasks on each processing core is controlled to smooth the overall energy consumption curve.

[0060] Specifically, the following technical implementation details are included: Task mapping function definition: Define the task scheduling mapping function. ,in, For a set of tasks (e.g., the control tasks in a task dependency graph). To process the core set. Mapping function. Indicates the task Scheduled to core This allows tasks to be executed in parallel within a multi-core environment.

[0061] Communication path planning and cost function: Communication path planning uses a communication cost function. Here, we introduce indicator functions. To indicate the task and tasks Whether they are assigned to different cores: When the two tasks are located on the same core, When located in different cores, . For the task and Data communication delay between them. Therefore, Indicates task With the task The cost incurred when communicating across cores (the cost is 0 if the tasks are located on the same core).

[0062] Task scheduling optimization objective: The objective of task scheduling is to minimize the overall communication cost. ,in, This is the set of task dependency edges. That is, while satisfying task dependencies, we should try to allocate dependent tasks to the same processing core (at this point, This reduces cross-core data transfer overhead by minimizing communication costs to zero. Minimization can reduce communication latency caused by tasks being executed on different cores, thereby improving the efficiency of task scheduling.

[0063] Energy consumption smoothing control objective: To smooth the overall energy consumption curve, it is necessary to control the power consumption fluctuations of multi-core systems during operation. This involves introducing time series analysis. Indicates the system at each time point Total power consumption Let be the average power consumption within the selected time window. The optimization objective of energy consumption smoothing is to minimize the variance of power consumption within the sliding window, i.e.: ,in, The length of the sliding window. This represents the start time of the window. By minimizing this power consumption variance, the fluctuation of power consumption over time can be effectively controlled, making the total power consumption curve smoother and avoiding sudden increases or decreases in power consumption caused by multiple cores running under high load simultaneously.

[0064] When conditions such as core degradation or resource constraints are detected, a dynamic computing power reconfiguration mechanism is triggered to automatically readjust the task dependency graph structure and repair the task scheduling scheme in real time to adapt to changes in available computing power.

[0065] In the above scheme, when a processing core degradation or resource limitation is detected (such as a decrease in available computing power, temperature / utilization exceeding the threshold, or a fault), a dynamic computing power reconstruction mechanism is triggered: failure / limited cores are removed in real time, the set of available cores is rebuilt, a subgraph is formed in the task dependency graph for the affected tasks and their adjacent dependencies, the edge weights (communication / timing overhead) are recalculated, and a new "task → core" mapping and execution sequence are generated to ensure that critical tasks are completed on schedule while taking into account communication latency and energy consumption smoothness.

[0066] In the above scheme, the trigger criterion function is as follows: when the available computing power of any processing core is lower than the threshold or any of the temperature, utilization rate, or error rate exceeds the threshold, reconstruction is triggered.

[0067] Available core update function: Remove cores that do not meet the threshold condition from the original core set to obtain a new available core set.

[0068] Affected subgraph extraction function: From the original task dependency graph, select the "task originally mapped to the exception core" and its predecessor / successor dependent tasks to form the task subgraph to be remapped and the corresponding set of dependency edges.

[0069] Joint optimization of remapping and temporal sequence (objective function + constraints): Objective: Minimize the total cost of cross-core communication plus the penalty for breach of deadline plus the penalty for energy consumption fluctuation.

[0070] Constraints: Dependency order must be satisfied, computing power / capacity per core must not exceed limits, and task start-end time must be consistent.

[0071] Output: The new "Task → Core" mapping and the start / end time of each task.

[0072] Graph repair and edge weight update: Update the cross-core edge weights on the original task graph with the new mapping results (the communication cost is counted for cross-core edges, and zero for edges within the same core), to obtain the repaired task dependency graph and executable timing.

[0073] The task completion status, real-time indicators, and energy consumption data during the execution phase are fed back to the reinforcement learning engine and heterogeneous perception module to update subsequent scheduling strategies and model parameters, thereby forming a closed loop of the entire process of perception, modeling, decision-making, scheduling, execution, and feedback.

[0074] During operation, the feedback module gathers task completion status, real-time indicators (whether there is a default / time margin) and energy consumption data from the execution phase, and feeds them back to the reinforcement learning engine and heterogeneous perception module to update the state representation, optimize the target weights and policy parameters, and then iteratively correct the scheduling strategy to achieve adaptive closed-loop optimization of "perception-modeling-decision-scheduling-execution-feedback".

[0075] The above plan includes: Real-time feedback function: Calculates the time margin of "deadline minus actual completion time" for each task; and calculates the "on-time completion rate" within the window.

[0076] Communication and energy consumption feedback function: Statistical window for "total cost of cross-core communication"; calculates the power consumption increment at adjacent time points and the power consumption variance within the window (used to characterize energy consumption stability).

[0077] State update function: It integrates the "previous system state + real-time / communication / energy consumption feedback + heterogeneous sensing module output (environmental parameters and operating state parameters)" into the reinforcement learning state input for the next step.

[0078] Policy parameter update function: The loss is constructed based on "multi-objective weighted reward", and the policy network and value network parameters are updated by gradient descent.

[0079] Dynamic weight adaptive rules: When the "real-time performance compliance rate is lower than the threshold", the "real-time performance weight" is increased; when the "power consumption deviates from the window average threshold", the "energy consumption stability weight" is increased; the remaining weights are reduced proportionally and kept to a total of 1, forming a closed-loop adaptive optimization of "perception-modeling-decision-scheduling-execution-feedback".

[0080] The heterogeneous sensing module is used to collect multi-source input parameters of the wind power main control system. These multi-source input parameters include two categories: environmental parameters and operating status parameters. Environmental parameters include at least wind speed, wind direction, ambient temperature, and grid load. Operating status parameters include at least rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and the utilization rate and temperature of each processing core of the main control multi-core controller. The parameter update cycle of the heterogeneous sensing module is dynamically adjusted according to wind conditions and system operating status. Specifically, corresponding update cycles are set for environmental parameters and operating status parameters respectively, and the update cycles are adaptively adjusted according to their rate of change to ensure the real-time performance and accuracy of the sensing data. The collected environmental parameters and operating status parameters are combined to form sensing data, which is then output to the reinforcement learning engine for subsequent scheduling decisions.

[0081] The task dependency graph is constructed using a directed acyclic graph model. Each node represents a wind power master control task and includes attribute information such as the task's execution cycle, deadline, and expected execution time. The directed edges between nodes represent the data or control dependencies between tasks. The scheduling input interface inputs the constructed task dependency graph and its node attribute information into the reinforcement learning scheduling engine as the basis for task scheduling decisions.

[0082] The reinforcement learning engine models the task scheduling process as a Markov decision process. Its state space includes information such as the ready task queue in the current task dependency graph, the running load status of each processing core, and the remaining time before the task deadline. The action space includes scheduling decisions such as selecting a target processing core for ready tasks or adjusting the execution priority of tasks. The reward and penalty function of the reinforcement learning engine comprehensively considers indicators such as real-time failure, communication latency, and energy consumption fluctuations. It gives positive rewards to scheduling behaviors that complete tasks on time, reduce inter-core communication latency, and stabilize the energy consumption curve, and imposes penalties on scheduling behaviors that exceed the deadline or introduce high communication latency and drastic energy consumption fluctuations, thereby guiding the scheduling strategy to converge in the direction of satisfying multi-objective optimization.

[0083] The multi-objective optimization function employs a weighted collaborative adjustment strategy for each sub-objective, including real-time performance, communication delay, and stable energy consumption. Specifically, it dynamically adjusts the weight coefficients of each sub-objective based on the operating conditions of the wind power main control system and the priority control objectives. When a deterioration in real-time performance is detected, the weight of the real-time performance objective is increased; when communication delay increases, the weight of the communication delay objective is increased; and when energy consumption fluctuations increase, the weight of the stable energy consumption objective is increased. This approach aims to balance the conflicts between various optimization objectives under different operating conditions and ensure the overall performance of the scheduling optimization.

[0084] For cross-processing core task data transmission, an inter-core communication path planning algorithm is used to optimize communication latency: tasks with direct data dependencies are prioritized to be assigned to processing cores with high-speed interconnect channels to shorten the inter-core communication distance and time; for tasks that must be executed on different processing cores and need to communicate with each other, the shortest communication path is planned and the communication timing is time-sharing to avoid bus contention conflicts, thereby reducing communication latency and bandwidth overhead.

[0085] A power consumption curve smoother is configured to smooth the power consumption curve of a multi-core controller in real time. The smoother monitors the instantaneous power consumption of each processing core and predicts the load change trend, dynamically adjusting the task scheduling rhythm or the working frequency of the processing core to avoid power consumption spikes caused by the simultaneous execution of high-power tasks. By fine-tuning and peak-shifting the task execution sequence, the overall power consumption curve over time becomes smoother, reducing the impact of power consumption fluctuations on system stability.

[0086] A fault-adaptive reconfiguration scheduling mechanism is set up. When a processing core is detected to have experienced performance degradation or failure, or when the system resource utilization rate exceeds a preset threshold, the reconfiguration scheduling mechanism is triggered. During the reconfiguration process, affected processing cores are removed or downgraded in real time, and the task dependency graph is adjusted. Tasks originally assigned to the affected cores are remapped and reassigned to other normally functioning cores, and an updated task scheduling scheme is dynamically generated. During the scheduling recovery process, critical control tasks are prioritized for migration and continued execution to ensure the continuity and stability of the wind power main control system control process.

[0087] When the wind farm control subsystem is under high load conditions, a special scheduling weight adjustment strategy is implemented. High load conditions include situations where extreme winds significantly increase the load on pitch control and yaw control tasks. In such cases, the weight or priority of key tasks such as pitch control and yaw control is increased in scheduling optimization to ensure that these key control tasks receive core processing resources with higher priority and shorten response time. Even if the weight of other objectives (such as energy consumption optimization) is temporarily reduced as a result, the safety and stability of wind turbine operation are prioritized.

[0088] The main control tasks of wind power include the control tasks of wind turbine subsystems such as blade pitch control, nacelle yaw control, generator power regulation, converter control, and braking system control. Each control task establishes corresponding dependency connections in the task dependency graph based on its functional logic and data interaction relationship.

[0089] The processing method provided in this application embodiment introduces parallel execution on different processing cores of a multi-core controller, and distributes each control task in the task dependency graph to be executed in parallel on different processing cores of the multi-core controller, which can greatly improve processing efficiency.

[0090] In the above embodiments, a method for cross-core task scheduling and energy efficiency coordination based on a multi-core controller is provided. Correspondingly, this application also provides a system for cross-core task scheduling and energy efficiency coordination based on a multi-core controller. The system for cross-core task scheduling and energy efficiency coordination based on a multi-core controller provided in this application can implement the above-described method for cross-core task scheduling and energy efficiency coordination based on a multi-core controller. This system can be implemented through software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the above methods.

[0091] Please refer to Figure 2 This illustration shows a schematic diagram of a cross-core task scheduling and energy efficiency coordination processing system based on a multi-core controller, provided by some embodiments of this application. Since the system embodiments are basically similar to the method embodiments, the description is relatively simple; relevant details can be found in the descriptions of the method embodiments. The system embodiments described below are merely illustrative.

[0092] like Figure 2 As shown, the processing system 200 based on a multi-core controller for cross-core task scheduling and energy efficiency coordination may include: The heterogeneous sensing module 201 is used to collect multi-source input parameters of the wind power main control system in real time. The multi-source input parameters include environmental parameters and operating status parameters. The first construction module 202 is used to construct the task dependency graph of the wind power main control task based on the graph structure modeling method; The generation module 203 is used to generate a task dependency model by taking each control task as a node and the dependency relationship between tasks as an edge. In the task dependency graph, each control task is taken as a node and the dependency relationship between tasks is taken as an edge. The second building module 204 is used to construct the state space and action space of reinforcement learning based on the dependency graph; Configuration module 205 is used to configure a reward and punishment mechanism including a multi-objective optimization function, wherein the optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication delay, and stabilizing the energy consumption function curve; The parallel execution module 206 is used to allocate each control task in the task dependency graph to different processing cores of the multi-core controller for parallel execution according to the task scheduling strategy corresponding to the reinforcement learning engine, so as to obtain the task completion status of the execution phase. Trigger module 207 is used to trigger the dynamic computing power reconfiguration mechanism when the current conditions are detected to meet the conditions of processing core degradation or resource limitation. The adjustment and repair module 208 is used to automatically readjust the graph structure of the task dependency graph based on the computing power dynamic reconstruction mechanism and to repair the task scheduling scheme in real time. Feedback module 209 is used to feed back the task completion status, real-time indicators and energy consumption data of the execution phase to the reinforcement learning engine and heterogeneous perception module to obtain feedback results. Based on the feedback results, the model parameters of the task scheduling strategy and the task dependency model are updated to obtain the updated task scheduling strategy and the updated task dependency model.

[0093] In some embodiments of this application, the processing system 200 based on a multi-core controller for cross-core task scheduling and energy efficiency coordination may further include: First acquisition module (in) Figure 2 (Not shown in the image) is used to acquire multi-source input parameters after real-time acquisition of multi-source input parameters of the wind power main control system. Environmental parameters include at least: wind speed, wind direction, ambient temperature, and grid load; operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and utilization rate and temperature of each processing core of the main control multi-core controller. The second acquisition module (in) Figure 2 (not shown in the image), used to obtain information based on wind conditions and system operating status; Dynamic adjustment module (in) Figure 2 (Not shown in the image), used to dynamically adjust the update cycle of each parameter in the multi-source input parameters according to wind conditions and system operating status.

[0094] In some embodiments of this application, the processing system 200 based on a multi-core controller for cross-core task scheduling and energy efficiency coordination may further include: The third acquisition module (in) Figure 2 (Not shown in the image) is used for: after constructing the task dependency graph of the wind power master control task using a graph structure modeling method, obtaining the task dependency graph, wherein the task dependency graph is constructed based on a directed acyclic graph model. In the task dependency graph, each node represents a wind power master control task and multiple attribute information corresponding to that wind power master control task. The multiple attribute information includes: the execution cycle, deadline, and expected execution time of the task; the directed edge between any two nodes in each node represents the data relationship or control dependency relationship between the tasks; and obtaining the node attribute information of each node in the task dependency graph. Processing module (in) Figure 2 (Not shown in the image), used to input the task dependency graph and the node attribute information of each node into the reinforcement learning engine.

[0095] In some embodiments of this application, the second construction module 204 is specifically used for: By using a reinforcement learning engine, the task scheduling process is transformed into a Markov decision process; The state space and action space of reinforcement learning are constructed based on the dependency graph. The state space includes: the queue of ready tasks in the current task dependency graph, the running load status of each processing core, and the remaining time before the task deadline. The action space includes: scheduling decisions, which are scheduling decisions used to select target processing cores for ready tasks or adjust the execution priority of tasks. Configuration module 205 is specifically used for: The configuration includes a reward and penalty mechanism for multi-objective optimization functions; Obtain the reward and punishment functions used by the reward and punishment mechanism; Based on multiple indicators, a reward and penalty function is used to give positive rewards to the first scheduling behavior that meets the first condition and to impose penalties on the second scheduling behavior that meets the second condition, so as to guide the scheduling strategy to converge in the direction of satisfying multi-objective optimization. The multiple indicators include: real-time failure, communication latency, and energy consumption fluctuation; the first condition includes: completing the task on time, or reducing inter-core communication latency, or stabilizing the energy consumption curve; the second condition includes: completing the task beyond the deadline, or introducing high communication latency, or drastic energy consumption fluctuation.

[0096] In some embodiments of this application, the processing system 200 based on a multi-core controller for cross-core task scheduling and energy efficiency coordination may further include: The fourth acquisition module (in) Figure 2 (Not shown in the image) is used to: obtain the operating status of the wind power main control system and obtain the priority control target after configuring a reward and punishment mechanism that includes a multi-objective optimization function; The fifth acquisition module (in) Figure 2 (not shown in the image), used to: acquire each sub-target, which includes: real-time performance, communication latency, and stable energy consumption; Dynamic adjustment module (in) Figure 2 (Not shown in the image), used to: dynamically adjust the weights of each sub-objective based on operating conditions and priority control objectives.

[0097] In some embodiments of this application, the dynamic adjustment module is specifically used for: If a deterioration in a real-time performance indicator is detected, the weight of that real-time performance indicator will be increased. If an increase in communication latency is detected, the weight corresponding to the communication latency indicator will be increased. If an increase in energy consumption fluctuation is detected, the weight of the energy consumption stability index will be increased.

[0098] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform the above-described cross-core task scheduling and energy efficiency coordination processing method based on a multi-core controller.

[0099] According to another embodiment, an electronic device is also provided, including a memory and a processor. The memory stores executable code, and when the processor executes the executable code, it implements the above-described processing method for cross-core task scheduling and energy efficiency coordination based on a multi-core controller.

[0100] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this application can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0101] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A cross-core task scheduling and energy efficiency coordination method based on a multi-core controller, characterized in that, The method includes: The heterogeneous sensing module collects multi-source input parameters of the wind power main control system in real time, including environmental parameters and operating status parameters. A task dependency graph for wind power main control tasks is constructed based on a graph structure modeling method, and a task dependency model is generated by taking each control task as a node and the dependency relationship between tasks as an edge. In the task dependency graph, each control task is taken as a node and the dependency relationship between tasks is taken as an edge. The state space and action space of reinforcement learning are constructed based on the dependency graph, and a reward and punishment mechanism including a multi-objective optimization function is configured. The optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication latency, and stabilizing the energy consumption function curve. According to the task scheduling strategy corresponding to the reinforcement learning engine, each control task in the task dependency graph is assigned to different processing cores of the multi-core controller for parallel execution, thereby obtaining the task completion status of the execution phase. When the current conditions are detected to meet the conditions of processing core degradation or resource limitation, the computing power dynamic reconstruction mechanism is triggered, and the graph structure of the task dependency graph is automatically readjusted based on the computing power dynamic reconstruction mechanism and the task scheduling scheme is repaired in real time. The task completion status, real-time indicators, and energy consumption data during the execution phase are fed back to the reinforcement learning engine and the heterogeneous perception module to obtain feedback results. Based on the feedback results, the model parameters of the task scheduling strategy and the task dependency model are updated to obtain the updated task scheduling strategy and the updated task dependency model.

2. The processing method according to claim 1, characterized in that, After acquiring multi-source input parameters of the wind power main control system in real time through the heterogeneous sensing module, the method further includes: The multi-source input parameters are obtained, and the environmental parameters include at least: wind speed, wind direction, ambient temperature, and power grid load; the operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and the utilization rate and temperature of each processing core of the main control multi-core controller. Obtain wind conditions and the operating status of the wind power main control system; The update cycle of each parameter in the multi-source input parameters is dynamically adjusted based on the wind conditions and the system operating status.

3. The processing method according to claim 1, characterized in that, After constructing the task dependency graph of the wind power master control task using the graph structure modeling method, the method further includes: Obtain the task dependency graph, which is constructed based on a directed acyclic graph model. In the task dependency graph, each node represents a wind power master control task and multiple attribute information corresponding to the wind power master control task. The multiple attribute information includes: the execution cycle, deadline, and expected execution time of the task. The directed edge between any two nodes in each node represents the data relationship or control dependency relationship between the tasks. Obtain the node attribute information of each node in the task dependency graph; The task dependency graph and the node attribute information of each node are input into the reinforcement learning engine for processing to obtain the processing result.

4. The processing method according to claim 1, characterized in that, The step of constructing the state space and action space for reinforcement learning based on the dependency graph, and configuring a reward and punishment mechanism including a multi-objective optimization function, includes: The task scheduling process is transformed into a Markov decision process through the reinforcement learning engine. The state space and action space of reinforcement learning are constructed based on the dependency graph. The state space includes: the queue of ready tasks in the current task dependency graph, the running load status of each processing core, and the remaining time of the task deadline. The action space includes: scheduling decisions, which are scheduling decisions used to select target processing cores for ready tasks or adjust the execution priority of tasks. The configuration includes a reward and penalty mechanism for multi-objective optimization functions; Obtain the reward and punishment function used by the reward and punishment mechanism; Based on the influence of multiple indicators, the reward and penalty function provides a positive reward for the first scheduling behavior that meets the first condition and imposes a penalty for the second scheduling behavior that meets the second condition, so as to guide the scheduling strategy to converge in the direction of satisfying multi-objective optimization. The multiple indicators include: real-time failure, communication latency, and energy consumption fluctuation; the first condition includes: completing the task on time, or reducing inter-core communication latency, or stabilizing the energy consumption curve; the second condition includes: completing the task beyond the deadline, or introducing high communication latency, or drastic energy consumption fluctuation.

5. The processing method according to claim 2, characterized in that, After the configuration includes a reward and penalty mechanism for a multi-objective optimization function, the method further includes: Obtain the operating status of the wind power main control system and the priority control targets; Obtain the sub-objectives, which include: real-time performance, communication latency, and stable energy consumption; Based on the operating conditions and the priority control objectives, the weights of each sub-objective are dynamically adjusted.

6. The processing method according to claim 5, characterized in that, The dynamic adjustment of the weights of each sub-objective based on the operating conditions and the priority control objectives includes: If a deterioration in a real-time performance indicator is detected, the weight of that real-time performance indicator will be increased. If an increase in communication latency is detected, the weight corresponding to the communication latency indicator will be increased. If an increase in energy consumption fluctuation is detected, the weight of the energy consumption stability index will be increased.

7. A processing system for cross-core task scheduling and energy efficiency coordination based on a multi-core controller, characterized in that, The system includes: The heterogeneous sensing module is used to collect multi-source input parameters of the wind power main control system in real time. The multi-source input parameters include environmental parameters and operating status parameters. The first building module is used to construct the task dependency graph of the wind power main control task based on the graph structure modeling method; The generation module is used to generate a task dependency model by taking each control task as a node and the dependencies between tasks as edges. In the task dependency graph, each control task is taken as a node and the dependencies between tasks are taken as edges. The second construction module is used to construct the state space and action space of reinforcement learning based on the dependency graph. The configuration module is used to configure a reward and punishment mechanism including a multi-objective optimization function, wherein the optimization objectives adopted by the multi-objective optimization function are: minimizing the real-time default rate, minimizing the inter-core communication delay, and stabilizing the energy consumption function curve; The parallel execution module is used to allocate each control task in the task dependency graph to different processing cores of the multi-core controller for parallel execution according to the task scheduling strategy corresponding to the reinforcement learning engine, so as to obtain the task completion status of the execution phase. The triggering module is used to trigger the dynamic reconfiguration mechanism of computing power when the current conditions are detected to meet the conditions of core degradation or resource limitation. The adjustment and repair module is used to automatically readjust the graph structure of the task dependency graph based on the computing power dynamic reconstruction mechanism and to repair the task scheduling scheme in real time. The feedback module is used to feed back the task completion status, real-time indicators and energy consumption data of the execution phase to the reinforcement learning engine and the heterogeneous perception module to obtain feedback results. Based on the feedback results, the model parameters of the task scheduling strategy and the task dependency model are updated to obtain the updated task scheduling strategy and the updated task dependency model.

8. The processing system according to claim 7, characterized in that, The system also includes: The first acquisition module is used to acquire the multi-source input parameters after real-time acquisition of the multi-source input parameters of the wind power main control system. The environmental parameters include at least: wind speed, wind direction, ambient temperature, and grid load. The operating status parameters include at least: rotor speed, generator speed and torque, blade pitch angle, nacelle yaw angle, DC bus voltage and current, and utilization rate and temperature of each processing core of the main control multi-core controller. The second acquisition module is used to acquire information based on wind conditions and system operating status. The dynamic adjustment module is used to dynamically adjust the update cycle of each parameter in the multi-source input parameters according to the wind conditions and the system operating status.

9. A computer-readable storage medium, characterized in that, It contains a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1 to 6.

10. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method according to any one of claims 1 to 6.

Images