Batch processing scheduling method for wind turbine design load case and related device

Abstract

This invention belongs to the field of wind turbine simulation and scheduling technology, and discloses a batch processing scheduling method and related device for wind turbine design load conditions. It aims to solve the technical problems of traditional wind power load simulation batch processing scheduling, such as coarse prioritization, low computing resource utilization, and difficulty in adapting to hybrid scheduling of local and cloud platforms. The invention first obtains a list of load conditions to be processed, and obtains the comprehensive importance of the conditions based on a multi-factor weighted model that integrates fault probability, damage severity, and computational complexity, thus completing dynamic priority ranking. Then, it determines the computing power scale by combining the resource status of the local cluster and the cloud platform, solves for the total number of parallel nodes using a constraint model, and uses a minimum load priority strategy to achieve condition allocation. Finally, it schedules each computing node to perform aerodynamic-structural coupled simulation calculations in parallel. This invention can accelerate the response speed of critical operating conditions, improve computing power utilization, and shorten the overall calculation cycle of wind turbine load analysis.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine simulation calculation and scheduling technology, and specifically relates to a batch processing scheduling method and related devices for wind turbine design load conditions. Background Technology

[0002] As large rotating mechanical equipment, wind turbines must strictly follow the IEC 61400-1 standard established by the International Electrotechnical Commission (IEC) in their design verification process. This standard defines a design load case (DLC) system covering various operating scenarios such as normal power generation, fault conditions, start-up / shutdown, and transportation / installation.

[0003] Taking IEC 61400-1 Ed.4 as an example, the standard specifies DLC types from DLC 1.1 to DLC 8.2. Each DLC type needs to cover multiple wind speed ranges (for illustration, typically 10 to 20 wind speed points from the cut-in wind speed to the cut-out wind speed) and multiple wind direction sectors. The total number of simulation cases required in actual engineering projects usually reaches hundreds or even thousands. The calculation of each DLC case is itself a computationally intensive task, involving aerodynamic load calculation (such as based on blade element momentum theory BEM or generalized dynamic wake GDW model), structural dynamic response solution (such as based on multibody dynamics or finite element method), and complex numerical simulation processes such as aerodynamic-structural coupling iteration. Taking the complete DLC simulation of a typical 3MW wind turbine as an example, the calculation time for a single case is about 30 to 90 minutes. Therefore, completing the serial simulation of all design load cases usually takes several weeks or even months.

[0004] In actual engineering projects, to meet the requirements of unit type certification and design optimization iteration, engineers often need to complete multiple rounds of design load simulation analysis in a short period of time. Existing traditional batch processing methods still have the following three limitations, including: First, the scheduling strategy is simplistic and crude. Specifically, existing batch processing schemes typically employ first-in-first-out (FIFO) or simple round-robin scheduling strategies, failing to consider the differences in importance between different DLC (Dynamic Continuous Load) conditions. Based on practical experience, the ultimate load of wind turbines usually occurs in a few critical conditions (such as DLC 1.3 extreme turbulence condition and DLC 6.1 50-year return period extreme wind speed condition). The simulation results of these critical conditions directly determine the turbine's structural size and cost, and should be prioritized. However, existing traditional scheduling methods treat all conditions with the same priority, resulting in the results of critical conditions often having to wait for a large number of routine conditions to be completed before they can be obtained, severely impacting the efficiency of design decisions.

[0005] Second, the utilization efficiency of computing resources is low. Specifically, traditional batch processing methods typically adopt a parallel mode with fixed resources, that is, a fixed number of computing nodes are preset to execute tasks in parallel. This lacks dynamic management and elastic scheduling capabilities for computing resources. This static resource allocation method has obvious shortcomings: on the one hand, when the task queue is long (such as when the batch processing task has just started), the fixed number of resources may be insufficient, causing some tasks to wait in the queue for too long; on the other hand, when the batch processing task is nearing its end (such as when most tasks have been completed and only a few tasks remain), a large number of computing resources are idle.

[0006] Third, there is a lack of integration capabilities with cloud computing platforms. Specifically, with the development of cloud computing technology, wind turbine simulation computing is gradually migrating from local clusters to hybrid cloud architectures. Traditional batch processing systems are usually designed for fixed cluster environments and cannot flexibly allocate tasks and schedule resources between local clusters and cloud platforms, thus limiting the scalability and real-time performance of computing resources. Summary of the Invention

[0007] The purpose of this invention is to provide a batch processing scheduling method and related apparatus for wind turbine design load conditions, to solve the technical defects of existing traditional wind power load simulation batch processing scheduling, such as coarse priority, low computing resource utilization, and inability to adapt to hybrid scheduling of local and cloud platforms. The technical solution disclosed in this invention improves the calculation response speed and overall resource utilization efficiency of key operating conditions, and can significantly shorten the overall calculation cycle of wind turbine design load analysis.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a batch processing scheduling method for wind turbine design load conditions, comprising the following steps: Step 1: Obtain the list of design load conditions for the wind turbine to be processed, and dynamically prioritize each condition in the list to obtain a priority ranking queue. When dynamically prioritizing each condition in the list, the comprehensive importance of each condition is used as the ranking index. The comprehensive importance is calculated by a multi-factor weighted model that takes into account the condition failure probability factor, the condition damage severity factor, and the condition calculation complexity factor. Step 2: Based on the priority sorting queue, dynamically allocate and load balance resources according to the current available computing resource status, and generate task allocation instructions. The current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the process of dynamically allocating and load balancing resources and generating task allocation instructions based on the current available computing resource status, the computing power scale is determined. Based on the computing power scale, the total number of parallel nodes is calculated according to the constraint model. Based on the total number of parallel nodes, a minimum load priority strategy is used to assign the design load condition to the computing node with the smallest estimated load value, generating task allocation instructions for the design load condition. Step 3: Based on the task allocation instructions, perform aerodynamic-structural coupling simulation calculations in parallel on each computing node to obtain the working condition simulation results.

[0009] A further improvement to the technical solution of the present invention is that the multi-factor weighted model is expressed as: ; In the formula, To assess overall importance, For operating condition index; This is the failure probability factor under operating conditions; This is a factor representing the severity of damage under operating conditions. The complexity factor for the operating conditions is calculated. , , These are the weighting coefficients for each factor.

[0010] A further improvement to the technical solution of this invention lies in the fact that, during the batch processing scheduling startup phase, the computational complexity factor of the working condition is... Determined through preliminary analysis; The pre-analysis includes: reading the simulation configuration parameters and wind field boundary conditions corresponding to each design load condition, estimating the computation time and hardware resource consumption required for aerodynamic-structural coupling simulation, and obtaining the computational complexity factor of the load condition.

[0011] A further improvement of the technical solution of the present invention lies in that, during the batch processing scheduling and execution phase, the complexity factor of the unexecuted work conditions is calculated based on the calculation results and time consumption statistics of the completed work conditions. Make corrections and updates; The corrected update expression is as follows: ; In the formula, For the first The computational complexity factor of the working conditions in the round of iteration; For the first The computational complexity factor of the working conditions in the round of iteration; The average calculation time for similar working conditions that have been observed to have been completed; This represents the maximum calculation time among all completed work conditions. This is a smoothing coefficient used to balance the weights between historical estimates and new observations.

[0012] A further improvement to the technical solution of this invention lies in the process of determining the computing power scale based on the current available computing resource status: When the number of design load conditions to be processed in the priority sorting queue exceeds the current idle computing power capacity, an expansion operation is triggered, and an application for additional computing nodes is made to the cloud platform. When the priority sorting queue is cleared and a preset cooldown period is maintained, and no new design load conditions are added, a scaling-down operation is triggered to release the temporary computing nodes of the cloud platform and maintain the local baseline computing capacity.

[0013] A further improvement to the technical solution of this invention lies in that, in the step of solving for the total number of parallel nodes based on the computing power scale and according to the constraint model, the constraint model is expressed as: ; In the formula, For a moment The total number of computing nodes allocated; Minimum number of nodes; The maximum number of nodes; This represents a fixed number of nodes in the local cluster. For a moment The number of waiting conditions in the queue; For a moment The number of idle nodes; This is the expansion factor; This is the reduction factor.

[0014] A further improvement of the technical solution of this invention lies in the step of assigning the design load condition to the calculation node with the smallest estimated load value using a minimum load priority strategy. The calculation expression for the estimated load is as follows: ; In the formula, To estimate the load on the computing nodes, For compute node index; The number of operating conditions allocated to the computing nodes; For the first The estimated calculation time for each working condition; This represents the total computing power of the computing nodes.

[0015] In a second aspect, the present invention provides a batch processing scheduling system for wind turbine design load conditions, comprising: The priority sorting module is used to obtain a list of wind turbine design load conditions to be processed and to dynamically prioritize each condition in the list to obtain a priority sorting queue. When dynamically prioritizing each condition in the list, the comprehensive importance of each condition is used as the sorting index. The comprehensive importance is calculated by a multi-factor weighted model that takes into account the condition failure probability factor, the condition damage severity factor, and the condition calculation complexity factor. The instruction generation module is used to dynamically allocate and load balance resources based on the priority sorting queue and the current available computing resource status, and generate task allocation instructions. The current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the process of generating task allocation instructions based on the current available computing resource status, the computing power scale is determined. Based on the computing power scale, the total number of parallel nodes is calculated according to a constraint model. Based on the total number of parallel nodes, a minimum load priority strategy is used to assign the design load condition to the computing node with the smallest estimated load value, generating the task allocation instructions for the design load condition. The parallel execution module is used to perform aerodynamic-structural coupling simulation calculations in parallel on each computing node based on the task allocation instructions, and obtain the working condition simulation results.

[0016] In a third aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements a batch scheduling method for wind turbine design load conditions as described in any one of the first aspects of the present invention.

[0017] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a batch scheduling method for wind turbine design load conditions as described in any one of the first aspects of the present invention.

[0018] Compared with the prior art, the present invention has the following beneficial effects: In existing technologies, most wind turbine load condition batch processing methods adopt static sequential scheduling, failing to differentiate the importance of load conditions. Furthermore, the fixed computing power scale cannot coordinate local cluster and cloud resources, easily leading to problems such as high-value load conditions queuing, cluster node load imbalance, wasted computing resources, and excessively long overall batch processing time. In view of these existing problems, this invention discloses a batch processing scheduling method for wind turbine design load conditions. This method dynamically prioritizes load conditions using a multi-factor comprehensive importance index, prioritizing critical load conditions. It dynamically determines the computing power scale based on the resource status of the local cluster and cloud platform, solves for the total number of parallel nodes using a constraint model, and completes load condition allocation using a minimum load priority strategy. Finally, it performs aerodynamic-structural coupling simulations in parallel on each computing node. This invention achieves dynamic management of load condition priorities, coordinated scheduling of local-cloud hybrid computing power, and node load balancing, prioritizing high-importance load conditions, improving the overall resource utilization of the cluster, and shortening the entire load condition batch processing cycle.

[0019] Existing technologies lack mathematical models for quantifying the overall importance of operating conditions. They rely solely on manual experience to roughly prioritize conditions, resulting in high subjectivity, unstable ranking, and an inability to balance safety risks, damage impacts, and simulation computational costs. This invention provides a weighted calculation formula for quantifying overall importance, integrating factors such as operating condition failure probability, damage severity, and computational complexity. By flexibly adjusting the weighting coefficients to adjust the influence ratio of these three types of indicators, it achieves an objective and quantitative assessment of operating condition importance. This provides a reproducible and quantifiable basis for dynamic priority ranking, improving the scientific rigor and stability of the ranking results.

[0020] Existing technologies lack methods for estimating the computational time of operating conditions at the initial stage of scheduling, and can only uniformly assume that the computational costs of all operating conditions are similar, resulting in the initial priority ranking deviating from the actual computing power requirements. This invention, during the batch processing startup phase, pre-analyzes and reads the operating condition configuration and wind field boundary conditions to estimate the time and hardware usage of the aerodynamic-structural coupling simulation, obtaining the initial operating condition computational complexity factor. This provides the basic input for the first round of comprehensive importance calculation, ensuring that the priority ranking during the scheduling startup phase closely matches the actual computing power requirements of the operating conditions and reducing initial scheduling decision deviations.

[0021] Traditional scheduling schemes use fixed and unchanging workload complexity parameters, which cannot continuously correct prediction deviations based on real simulation data. As batch processing continues to execute, priority prediction errors accumulate. This invention utilizes the actual time consumption statistics of completed workloads during the scheduling execution phase, and continuously updates the computational complexity factor of unexecuted workloads through a smooth iterative formula. By integrating historical predictions and measured operational data, it continuously reduces the prediction deviation of workload computing power, dynamically optimizes the priority ranking accuracy of subsequent workloads, and forms a self-optimizing scheduling closed-loop system.

[0022] Existing fixed-power scheduling models cannot adapt to tidal loads in the workload queue. During high loads, insufficient computing power leads to task backlog, while during low loads, cloud resources are occupied for extended periods, increasing simulation costs. This invention sets expansion and contraction trigger conditions based on the comparison between the number of workload queues and idle computing power. When the load is too high, it requests node expansion from the cloud platform; when the queue is cleared and there are no new tasks during the cooling-off period, it releases cloud nodes and retains local baseline computing power. This invention enables the computing power scale to dynamically scale with the workload, improving parallel processing capabilities during peak periods and saving cloud computing power costs during off-peak periods, balancing processing efficiency and resource consumption.

[0023] Simply relying on linearly increasing or decreasing the number of nodes based on the queue length can easily lead to problems such as exceeding the node limit and frequent jittering and oscillation, lacking upper and lower boundary constraints. This invention provides a parallel node total number constraint model, introducing a maximum node upper limit and a local baseline minimum node lower limit, and combining the number of queue waiting conditions, the number of idle nodes, the expansion coefficient, and the shrinkage coefficient to solve for the node number in real time. This invention constrains the node size range, avoiding unlimited expansion or continuous frequent fluctuations in the number of nodes, and ensuring the stable operation of the cluster.

[0024] Conventional load balancing strategies simply allocate tasks based on the current resource usage of nodes, without considering the subsequent computational overhead of new workloads, which can easily lead to long-term load imbalances. This invention integrates the estimated total time of already allocated workloads within a node with the node's computing power to quantify the estimated load on each node, continuously assigning new workloads to the node with the lowest load. This invention's allocation decision considers both the current load and the total number of tasks to be executed, achieving long-term global load balancing and avoiding situations where some nodes experience task backlog while others remain idle. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating a batch scheduling method for wind turbine design load conditions in an embodiment of the present invention. Figure 2 This is a schematic diagram of the architecture of a three-layer scheduling pyramid in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the priority dynamic sorting algorithm in an embodiment of the present invention; Figure 4 This is a schematic diagram of the resource elastic scaling mechanism in an embodiment of the present invention; Figure 5This is a bar chart illustrating the comparison of computational efficiency in an embodiment of the present invention; Figure 6 This is a schematic diagram of a batch processing scheduling system for wind turbine design load conditions in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of protection of the present invention, but rather to illustrate specific selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other.

[0029] While existing technologies offer scheduling schemes for high-performance computing tasks, such as Slurm and PBS, as well as some domain-specific batch optimization methods, these general-purpose scheduling schemes are not optimized for the specific needs of wind turbine DLC simulation. They lack understanding of the correlations and importance differences between DLC operating conditions, and do not possess the ability to perceive the characteristics of aerodynamic-structural coupling simulation tasks (such as differences in computational loads between different operating conditions, and data dependencies between certain operating conditions). Therefore, a dedicated batch scheduling method for wind turbine design load conditions is urgently needed to comprehensively consider factors such as operating condition importance, dynamic allocation of computational resources, and multi-level scheduling coordination, thereby significantly shortening the total computation cycle of large-scale DLC simulations.

[0030] This invention discloses a three-layer scheduling pyramid batch processing method for large-scale design load conditions defined by the IEC 61400-1 standard. This method belongs to the interdisciplinary technology direction of high-performance computing scheduling and wind turbine design verification. By constructing a three-layer scheduling architecture of "load condition priority sorting, dynamic resource allocation, and parallel execution of calculations", and combining a dynamic priority sorting algorithm based on load condition importance and a computing resource elastic scaling mechanism, this method can achieve efficient batch processing scheduling for large-scale DLC simulation, which can significantly shorten the calculation cycle of wind turbine design load analysis.

[0031] Please see Figure 1 The present invention provides a batch processing scheduling method for wind turbine design load conditions, which specifically includes the following steps: Step 1: Obtain the list of design load conditions for the wind turbine to be processed, and dynamically prioritize each condition to obtain a priority ranking queue. When dynamically prioritizing each condition, the overall importance of each condition is used as the ranking index. The overall importance is calculated using a multi-factor weighted model that takes into account the condition's failure probability factor, damage severity factor, and computational complexity. Step 2: Based on the priority sorting queue, dynamically allocate and load balance resources according to the current available computing resource status, and generate task allocation instructions for the design load conditions. Specifically, the current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the dynamic allocation and load balancing of resources based on the current available computing resource status, the computing power scale is determined based on the current available computing resource status. Based on the computing power scale, the total number of parallel nodes is solved according to the constraint model. Based on the total number of parallel nodes, the design load condition tasks are allocated using a minimum load priority strategy, that is, the design load conditions to be scheduled are assigned to the computing node with the smallest real-time load value, and the corresponding design load condition task allocation instructions are generated. Step 3: Based on the generated task allocation instructions, perform aerodynamic-structural coupling simulation calculations in parallel on each computing node. In a further preferred technical solution, node operation data is continuously collected during simulation execution and summarized to form operating condition statistics; after the simulation calculation is completed, the operating condition simulation results are output, and the operating condition statistics are simultaneously transmitted back to correct the operating condition calculation complexity and iteratively optimize the priority sorting and resource allocation logic.

[0032] The technical solution disclosed in this invention constructs a comprehensive importance index based on operating condition faults, damage levels, and computational complexity to achieve dynamic priority ranking of operating conditions, prioritizing the operation of high-value load conditions; it dynamically adjusts the computing power scale based on the current available computing power status, and completes elastic resource scheduling and load balancing by relying on constraint models and minimum load priority strategies; it transmits back running statistics after parallel simulation execution to continuously correct the computational complexity of operating conditions, forming a closed-loop adaptive scheduling; it overcomes the shortcomings of traditional batch processing static scheduling and fixed computing power, takes into account the importance of operating conditions and simulation time, fully utilizes the hybrid computing power of local and cloud, improves the overall operating efficiency of load condition batch processing, and shortens the response cycle of important operating conditions.

[0033] In this embodiment of the invention, a three-layer scheduling pyramid architecture is proposed, consisting of three layers from top to bottom: the top layer (working condition priority sorting layer), the middle layer (resource dynamic allocation layer), and the bottom layer (computation parallel execution layer). The layers form a closed-loop scheduling system through the downlink transmission of scheduling instructions and the uplink feedback of execution.

[0034] Please see Figure 2 , Figure 2 The overall design of the three-layer scheduling pyramid in the method of this invention is specifically illustrated. The top layer is the "Work Case Priority Ranking" layer, responsible for the importance assessment and priority calculation of DLC work cases; the middle layer is the "Dynamic Resource Allocation" layer, responsible for the elastic scaling and load balancing of computing nodes; and the bottom layer is the "Parallel Computing Execution" layer, responsible for the aerodynamic-structural coupling simulation calculation of multiple nodes. The arrows between the layers indicate the downlink transmission of scheduling instructions and the uplink feedback of execution, forming a closed-loop scheduling system. The feedback path on the left indicates the execution feedback information flow from the bottom layer to the top layer.

[0035] Specifically, the top layer is the load condition priority sorting layer. This layer is responsible for receiving the list of DLC load conditions to be processed and dynamically prioritizing all load conditions based on factors such as the design importance of each load condition, data dependency with completed load conditions, and expected computational complexity. The sorting process adopts a multi-factor weighted model based on load condition importance, comprehensively considering multiple dimensions such as load condition type weight, wind speed range representativeness, and expected ultimate load contribution. The sorting result is a load condition execution queue arranged in descending order of priority, which serves as the basis for lower-level scheduling decisions. The middle layer is the dynamic resource allocation layer. This layer receives the priority sorting queue output from the top layer and performs dynamic resource allocation and load balancing based on the current available computing resource status (including the number and status of local cluster nodes, cloud platform resource availability, etc.). This layer maintains a real-time status table of a computing resource pool, recording the CPU / GPU utilization, memory usage, and task execution status of each computing node. When the number of waiting conditions in the queue exceeds the current number of idle resources, an elastic expansion mechanism is triggered to request additional computing resources from the cloud platform. After the queue is cleared, a cooling-off period (usually 5-15 minutes) is maintained. If there are no new tasks, the temporary resources of the cloud platform are released, and only the local baseline capacity is retained. The bottom layer is the parallel execution layer. This layer receives specific computing task instructions allocated by the middle layer and performs aerodynamic-structural coupling simulation calculations in parallel on each computing node. Each computing node independently runs a simulation solver instance, executes the complete DLC simulation process (including wind field generation, aerodynamic load calculation, structural dynamic response solving, post-processing, etc.) according to the received condition parameters, and sends the calculation results and performance statistics back to the scheduling system.

[0036] In this embodiment of the invention, the information flow mechanism in the three-layer architecture is as follows: scheduling instructions are transmitted and refined layer by layer in the direction of "top layer → middle layer → bottom layer". The top layer outputs a task priority queue to the middle layer, which then formulates a specific task allocation scheme and issues it to the bottom layer nodes. Feedback information is summarized and reported layer by layer in the direction of "bottom layer → middle layer → top layer". The bottom layer reports back the calculation status and completion time of each task, the middle layer summarizes and evaluates the current resource utilization efficiency, and the top layer dynamically adjusts the priority of subsequent tasks based on the results of completed tasks and the dependencies of remaining tasks. This bidirectional information flow ensures that the scheduling system can adapt to dynamic changes during the calculation process.

[0037] Please see Figure 3 , Figure 3 This demonstrates the complete process of a priority ranking algorithm based on the importance of work conditions. Starting from the work condition input, the algorithm sequentially goes through importance assessment, priority calculation, sorting, allocation to available resources, and parallel execution by multiple nodes. The judgment node checks for any incomplete work conditions; if any are found, the priority is adjusted through the feedback path and the algorithm re-enters the assessment process; otherwise, the batch processing is completed.

[0038] To address the varying importance of different DLC operating conditions for wind turbines, this invention designs a dynamic priority ranking algorithm based on multi-factor weighting. This algorithm defines a comprehensive importance index for each operating condition to be processed, calculated as follows: ; In the formula, For the first The overall importance of each working condition is normalized to interval; The failure probability factor for the operating condition represents the probability of this operating condition occurring within the unit's life cycle. It takes a smaller value for normal power generation conditions (DLC 1.x) and a larger value for extreme event conditions (DLC 6.x). The severity factor of the working condition characterizes the severity of the structural damage consequences that the working condition may cause. The computational complexity estimate for the operating condition is obtained through rapid evaluation during the pre-analysis phase, and the value range is [range missing]. The more complex the working condition, the closer the value is to 1, that is... The smaller the value, the more likely the less complex critical operating conditions can be scheduled. , , The weight coefficients for each factor are, for example, taken as follows: , , It can also be adjusted according to project requirements.

[0039] In this embodiment of the invention, the dynamic adjustment strategy of the sorting algorithm is reflected in the following two aspects: Initial Priority Calculation: When the batch processing task starts, based on historical experience and pre-analysis results, an initial importance assessment is performed on all DLC scenarios, and the priority of each scenario is calculated. The values ​​are sorted in descending order to form the initial execution queue.

[0040] Dynamic feedback adjustment: During batch processing, based on the calculation results and time statistics of completed tasks, adjustments are made to tasks that have not yet been executed. The value is corrected and updated.

[0041] Specifically, the following update rules are adopted: ; In the formula, For the first Complexity estimates in round iterations; For the first Complexity estimates in round iterations; The average calculation time for similar working conditions that have been observed to have been completed; This represents the maximum calculation time among all completed work conditions. This is a smoothing coefficient, with a range of values. This is used to balance the weights between historical estimates and new observations.

[0042] Please see Figure 4 , Figure 4 The resource elastic scaling mechanism is demonstrated. With time on the horizontal axis and the amount of computing resources on the vertical axis, the dynamic behavior of the elastic scaling mechanism is shown. Figure 4 The text specifically illustrates a typical scheduling cycle from t0 to t4: initially (t0-t1), the load is low, maintaining the baseline capacity N. min During periods of increased load, elastic scaling is triggered, and the increase in resource quantity follows a certain pattern. The growth pattern; the peak steady-state phase maintains the maximum capacity N. max Nearby; after the load decreases, elastic scaling is triggered, gradually releasing cloud platform resources to restore baseline capacity.

[0043] To adapt to workload fluctuations during batch processing, this embodiment of the invention designs an elastic scaling mechanism for computing resources, supporting hybrid scheduling of local cluster nodes and cloud computing platforms.

[0044] The dynamic control of resource quantity adopts the following model: ; In the formula, For a moment The total number of computing nodes allocated; This represents the minimum number of nodes (i.e., the baseline capacity of the local cluster). The maximum number of nodes (subject to budget and cloud platform quotas); This represents a fixed number of nodes in the local cluster. For a moment The number of waiting conditions in the queue; For a moment The number of idle nodes; Expansion factor (range of values) (This represents the number of incremental nodes waiting for each condition to be triggered). The shrinkage factor (range of values) (used to control the rate of volume reduction).

[0045] Specifically, elastic scaling strategies include the following three modes: Basic operating mode: when the queue length At that time, only local cluster resources are used to maintain This node configuration is suitable for regular workloads or the initial / final stages of batch processing.

[0046] Elastic expansion mode: when the queue length And if the time exceeds a preset threshold window (e.g., 3 minutes), elastic scaling is triggered. The system then... The system calculates the required number of additional nodes and initiates a resource request to the cloud platform. The initialization of cloud platform nodes includes steps such as simulation environment deployment and license verification. After expansion, the system will schedule some low-priority tasks to be executed on cloud platform nodes.

[0047] Elastic scaling down mode: When the queue is completely emptied, or If the cooldown period continues beyond the preset timeout (e.g., 10 minutes), elastic scaling down will be triggered to gradually release the computing nodes of the cloud platform while preserving the baseline capacity of the local cluster.

[0048] Resource allocation employs an improved Least Loaded First (LLF) strategy, namely: ; In the formula, For the first Estimated load on each computing node; The number of operating conditions assigned to this node; For the first The estimated calculation time for each working condition; For nodes Total computing power (measured in standard computing units). The scheduler always assigns new tasks to... The node with the smallest value achieves global load balancing.

[0049] In this embodiment of the invention, based on the above three-layer architecture and core algorithm, the complete workflow of the batch processing method is as follows: Step 1: Initialization Phase. The system loads the DLC operating condition list (reading parameters such as operating condition type, wind speed range, wind direction, and sector from the IEC standard DLC data table), performs pre-analysis to obtain initial estimates of the computational complexity for each operating condition; simultaneously, it initializes the computing resource pool and configures the connection parameters between local cluster nodes and the cloud platform.

[0050] Step Two: Priority Ranking Phase (Top Level). Execute an importance-based priority ranking algorithm on all pending work conditions to calculate the overall importance of each work condition. Arrange them in descending order to form the initial execution queue. ,in .

[0051] Step 3: Resource Allocation Phase (Middle Layer). Check the current available computing resource status and allocate resources based on queue length. Calculate the required number of nodes and trigger elastic scaling if necessary. Use an LLF (Limited-Leadership Flow) strategy to prioritize nodes at the head of the queue. Each operating condition is assigned to each computing node.

[0052] Step 4: Parallel Execution Phase (Lower Layer). Each computing node independently executes its assigned DLC simulation task. Each node runs the complete simulation process and, upon completion, sends the result files (load timing, ultimate load statistics, etc.) and performance metrics (actual computation time, peak memory usage, etc.) back to the scheduling system.

[0053] Step 5: Feedback and Update Phase. The system collects the execution results returned by each node and updates the complexity estimate of the working condition. Reassess the priority of the remaining operating conditions. If the queue is not empty, proceed to step two; if the queue is empty, proceed to the scaling-down assessment process.

[0054] Step Six: Convergence Determination. Once the execution results for all scenarios have been returned and verified, the batch processing task ends. The system records the efficiency statistics for this scheduling (total time, resource utilization, speedup ratio, etc.) as a reference for subsequent scheduling optimization.

[0055] The technical solution disclosed in this invention can significantly shorten the computation cycle, improve the response time of critical operating conditions, optimize the utilization efficiency of computing resources, and has good platform compatibility. Specifically, this invention uses a three-layer scheduling architecture and a priority sorting algorithm to prioritize the allocation of critical operating conditions to available resources, eliminating the invalid waiting time of critical operating conditions waiting for regular operating conditions to complete in traditional batch processing; at the same time, it uses an elastic scaling mechanism to automatically expand capacity during computationally intensive phases, making full use of computing resources. Experiments show that, taking a scenario with approximately 600 operating conditions from IEC standard DLC 1.1 to 8.2 as an example, the total computation cycle of this method can be shortened to less than 40% of the traditional serial method, a reduction of more than 60%. The priority dynamic sorting algorithm of this invention based on the importance of operating conditions ensures that the critical operating conditions with the greatest impact on design decisions (such as DLC 1.3, DLC 6.1, etc.) receive computing resources first and return results earliest, enabling design engineers to obtain key design feedback in the early stages of batch processing tasks, significantly shortening the design iteration cycle. The elastic scaling mechanism of this invention enables on-demand allocation of computing resources, automatically scaling up during peak load periods and automatically scaling down during off-peak periods, avoiding the dilemma of insufficient or wasted resources in the traditional fixed-resource model. Overall resource utilization is improved by approximately 20%-35% compared to the fixed-node model. This invention supports hybrid scheduling of local computing clusters and mainstream cloud computing platforms (such as AWS, Alibaba Cloud, Tencent Cloud, etc.), ensuring both data security (sensitive simulation data remains on the local cluster) and elastic scaling capabilities (coping with computing peaks through the cloud platform).

[0056] In a specific embodiment of the present invention, the specific implementation of the present invention is illustrated by taking the batch simulation of all design load conditions from DLC 1.1 to DLC 8.2 as defined by the IEC 61400-1 standard as an example.

[0057] Example Background: During the design and certification phase of a certain type of 3.5MW wind turbine generator, load simulation analysis for a total of 586 DLC (Digital Literary and Computer Computation) scenarios was required. Each scenario involved three main simulation stages: wind field generation, blade element momentum theory aerodynamic calculation, and multibody dynamics structural response solution. A local high-performance computing cluster with 16 computing nodes (each node with 32 CPU cores and 256GB of memory) was configured on-site, along with a cloud platform elastic computing resource pool (expandable up to 64 nodes).

[0058] The implementation steps of this invention are as follows: (1) Initialization phase. Import 586 operating condition parameters from the IEC standard DLC data table, including operating condition type labels (DLC 1.1 to DLC 8.2), wind speed values, wind direction angles, eddy current levels, and power grid fault types. Perform pre-analysis: run a 1-minute aerodynamic pre-analysis for each operating condition to estimate the aerodynamic computational complexity; label fault probability factors and severity factors based on operating condition types.

[0059] (2) Priority sorting. Using... The default weights are reorganized to calculate the overall importance of all 586 operating conditions. The ranking results show that the DLC 6.1 (once-in-50-year extreme wind speed) and DLC 1.3 (extreme turbulence) series of operating conditions are ranked first, while the DLC 8.1 (transportation) series is ranked last. This forms a descending priority queue.

[0060] (3) Resource Allocation. At startup, 16 local nodes are put into use, with a queue length of 586 >> 16, triggering elastic scaling. The system allocates resources according to... Calculate the required number of nodes (take) An additional 48 nodes were requested on the cloud platform, bringing the total to 64 computing nodes that can run in parallel.

[0061] (4) Parallel execution and dynamic adjustment. The top 64 highest priority tasks in the queue are allocated to each computing node. After each task is completed, the results and time consumption statistics are returned, and the complexity estimate is updated. When the queue length drops below 32, gradually release the cloud platform nodes, eventually reverting to the local 16-node baseline configuration.

[0062] (5) Results summary. The total calculation time for all 586 operating conditions was 178 hours (approximately 7.4 days), which is 65.8% shorter than the traditional serial method (approximately 520 hours, approximately 21.7 days). The results of critical operating conditions DLC 6.1 and DLC 1.3 were returned within 4 hours after the task was started, providing timely critical design feedback to design engineers.

[0063] Please see Figure 5 The computational efficiency of three batch processing methods is compared with the relative computation cycle as the vertical axis. Traditional serial batch processing (100%) is used as the baseline; simple parallel batch processing (about 65%) is an improvement over serial processing, but is limited by the fixed resource mode; the three-layer scheduling pyramid method of this invention (≤40%) achieves optimal efficiency through priority sorting and elastic scaling, and the computation cycle is shortened by more than 60% compared with the traditional method.

[0064] This invention discloses a three-layer scheduling pyramid architecture, the first of its kind for wind turbine DLC simulation. Through the coordinated operation of the top layer (priority ranking), middle layer (resource allocation), and bottom layer (parallel execution), a closed-loop scheduling system is formed, enabling efficient organization and management of large-scale simulation tasks. Based on a dynamic priority ranking algorithm for operating condition importance, this invention designs a multi-factor weighted model to quantitatively evaluate the characteristics of DLC operating conditions. Combined with a dynamic feedback adjustment mechanism, the scheduling system can adaptively optimize the execution order of operating conditions. This invention proposes a flexible resource control model based on queue length, supporting hybrid scheduling of local clusters and cloud platforms, achieving on-demand allocation of computing resources. This invention's dedicated scheduling method system for DLC differs from general job scheduling systems (such as Slurm and PBS). This invention deeply optimizes for the specific needs of wind turbine DLC simulation (differences in operating condition importance, unbalanced computational load, aerodynamic-structural coupling characteristics, etc.), forming a dedicated batch processing scheduling method.

[0065] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0066] Please see Figure 6 In this embodiment of the invention, a batch processing scheduling system for wind turbine design load conditions is provided, comprising: The priority sorting module is used to obtain a list of wind turbine design load conditions to be processed and to dynamically prioritize each condition in the list to obtain a priority sorting queue. When dynamically prioritizing each condition in the list, the comprehensive importance of each condition is used as the sorting index. The comprehensive importance is calculated by a multi-factor weighted model that takes into account the condition failure probability factor, the condition damage severity factor, and the condition calculation complexity factor. The instruction generation module is used to dynamically allocate and load balance resources based on the priority sorting queue and the current available computing resource status, and generate task allocation instructions. The current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the process of generating task allocation instructions based on the current available computing resource status, the computing power scale is determined. Based on the computing power scale, the total number of parallel nodes is calculated according to a constraint model. Based on the total number of parallel nodes, a minimum load priority strategy is used to assign the design load condition to the computing node with the smallest estimated load value, generating the task allocation instructions for the design load condition. The parallel execution module is used to perform aerodynamic-structural coupling simulation calculations in parallel on each computing node based on the task allocation instructions, and obtain the working condition simulation results.

[0067] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from a computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to execute batch scheduling methods for wind turbine design load conditions.

[0068] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the batch scheduling method for wind turbine design load conditions in the above embodiments.

[0069] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0070] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0071] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A batch processing scheduling method for wind turbine design load conditions, characterized in that, Includes the following steps: Step 1: Obtain the list of design load conditions for the wind turbine to be processed, and dynamically prioritize each condition in the list to obtain a priority ranking queue. When dynamically prioritizing each condition in the list, the comprehensive importance of each condition is used as the ranking index. The comprehensive importance is calculated by a multi-factor weighted model that takes into account the condition failure probability factor, the condition damage severity factor, and the condition calculation complexity factor. Step 2: Based on the priority sorting queue, dynamically allocate and load balance resources according to the current available computing resource status, and generate task allocation instructions. The current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the process of dynamically allocating and load balancing resources and generating task allocation instructions based on the current available computing resource status, the computing power scale is determined. Based on the computing power scale, the total number of parallel nodes is calculated according to the constraint model. Based on the total number of parallel nodes, a minimum load priority strategy is used to assign the design load condition to the computing node with the smallest estimated load value, generating task allocation instructions for the design load condition. Step 3: Based on the task allocation instructions, perform aerodynamic-structural coupling simulation calculations in parallel on each computing node to obtain the working condition simulation results.

2. The batch processing scheduling method for wind turbine design load conditions according to claim 1, characterized in that, The multi-factor weighted model is expressed as follows: ； In the formula, To assess overall importance, For operating condition index; This is the failure probability factor under operating conditions; This is a factor representing the severity of damage under operating conditions. The complexity factor for the operating conditions is calculated. , , These are the weighting coefficients for each factor.

3. The batch processing scheduling method for wind turbine design load conditions according to claim 2, characterized in that, During the batch processing scheduling startup phase, the computational complexity factor of the working condition is... Determined through preliminary analysis; The pre-analysis includes: reading the simulation configuration parameters and wind field boundary conditions corresponding to each design load condition, estimating the computation time and hardware resource consumption required for aerodynamic-structural coupling simulation, and obtaining the computational complexity factor of the load condition.

4. The batch processing scheduling method for wind turbine design load conditions according to claim 2, characterized in that, During the batch processing scheduling and execution phase, based on the calculation results and time consumption statistics of completed tasks, the complexity factor of tasks that have not yet been executed is calculated. Make corrections and updates; The corrected update expression is as follows: ； In the formula, For the first The computational complexity factor of the working conditions in the round of iteration; For the first The computational complexity factor of the working conditions in the round of iteration; The average calculation time for similar working conditions that have been observed to have been completed; This represents the maximum calculation time among all completed work conditions. This is a smoothing coefficient used to balance the weights between historical estimates and new observations.

5. The batch processing scheduling method for wind turbine design load conditions according to claim 1, characterized in that, In determining the computing power scale based on the current state of available computing resources: When the number of design load conditions to be processed in the priority sorting queue exceeds the current idle computing power capacity, an expansion operation is triggered, and an application for additional computing nodes is made to the cloud platform. When the priority sorting queue is cleared and a preset cooldown period is maintained, and no new design load conditions are added, a scaling-down operation is triggered to release the temporary computing nodes of the cloud platform and maintain the local baseline computing capacity.

6. The batch processing scheduling method for wind turbine design load conditions according to claim 1, characterized in that, Based on the aforementioned computing power scale, in the step of solving for the total number of parallel nodes according to the constraint model, the constraint model is expressed as: ； In the formula, For a moment The total number of computing nodes allocated; Minimum number of nodes; The maximum number of nodes; This represents a fixed number of nodes in the local cluster. For a moment The number of waiting conditions in the queue; For a moment The number of idle nodes; This is the expansion factor; This is the reduction factor.

7. The batch processing scheduling method for wind turbine design load conditions according to claim 1, characterized in that, In the step of assigning the design load case to the calculation node with the smallest estimated load value using the minimum load priority strategy, the calculation expression for the estimated load is as follows: ； In the formula, To estimate the load on the computing nodes, For compute node indexes; The number of operating conditions allocated to the computing nodes; For the first The estimated calculation time for each working condition; This represents the total computing power of the computing nodes.

8. A batch processing and scheduling system for wind turbine design load conditions, characterized in that, include: The priority sorting module is used to obtain a list of wind turbine design load conditions to be processed and to dynamically prioritize each condition in the list to obtain a priority sorting queue. When dynamically prioritizing each condition in the list, the comprehensive importance of each condition is used as the sorting index. The comprehensive importance is calculated by a multi-factor weighted model that takes into account the condition failure probability factor, the condition damage severity factor, and the condition calculation complexity factor. The instruction generation module is used to dynamically allocate and load balance resources based on the priority sorting queue and the current available computing resource status, and generate task allocation instructions. The current available computing resource status includes: the number of local cluster nodes, their running status, and the availability of cloud platform resources. During the process of generating task allocation instructions based on the current available computing resource status, the computing power scale is determined. Based on the computing power scale, the total number of parallel nodes is calculated according to a constraint model. Based on the total number of parallel nodes, a minimum load priority strategy is used to assign the design load condition to the computing node with the smallest estimated load value, generating the task allocation instructions for the design load condition. The parallel execution module is used to perform aerodynamic-structural coupling simulation calculations in parallel on each computing node based on the task allocation instructions, and obtain the working condition simulation results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the batch processing scheduling method for wind turbine design load conditions as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the batch processing scheduling method for wind turbine design load conditions as described in any one of claims 1 to 7.

Images