Computing power task scheduling method and device based on new energy consumption and storage medium

Abstract

The invention relates to a computing power task scheduling method and device based on new energy consumption and a storage medium. The method comprises the steps that a residual power output prediction curve is established according to historical output power data of at least one new energy power generation node and other load data taking power from the new energy power generation node, and the residual power output prediction curve is used for indicating residual output power data, which can be used for scheduling, of each new energy power generation node in a future preset time period; obtaining a task scheduling strategy of at least one computing power task based on the residual power output prediction curve; and based on the task scheduling strategy, controlling the computing power resource to execute at least one computing power task. According to the embodiment of the invention, the structural limitation of the load rigidity of the computing power center can be broken through, the computing power center side is taken as a leading view angle, the flexible scheduling of the computing power task facing the distributed new energy local consumption scene is realized, the light abandoning rate and the peak regulation pressure of the power grid are reduced, and the energy efficiency level and the green computing power of the operation of the computing power center are remarkably improved.

Description

Technical Field

[0001] This disclosure relates to the field of green energy technology, and in particular to a computing power task scheduling method, device and storage medium based on new energy consumption. Background Technology

[0002] Renewable energy is developing rapidly worldwide, with photovoltaic and wind power becoming crucial components of the power system. Numerous distributed renewable energy systems are deployed in urban parks, industrial plants, school rooftops, and other areas, creating abundant and dispersed power sources. However, due to their inherent intermittency and instability, these systems often experience a mismatch between power generation capacity and load demand, leading to frequent issues like wind and solar power curtailment. This not only wastes usable green energy resources but also increases the grid's peak-shaving and frequency regulation burden.

[0003] Meanwhile, with the rapid rise of emerging industries such as artificial intelligence, big data, and digital twins, computing centers, as the infrastructure supporting these technologies, have become typical representatives of new high-energy-consuming loads. Large computing centers often consume hundreds of millions of kilowatt-hours of electricity annually, and their power load is highly continuous, intensive, and rigid, making it difficult to flexibly adjust according to external power supply conditions. This operational characteristic means that computing centers have almost no ability to follow the fluctuations in renewable energy output, failing to provide effective support for the local consumption of renewable energy. Furthermore, the continuous high-load operation of computing centers during peak hours further exacerbates the peak pressure on the regional power distribution network, increasing the risk of grid failures and power supply bottlenecks.

[0004] Therefore, in response to the three major practical dilemmas faced in the current scenario: (1) the mismatch between new energy power generation and local consumption capacity, and the prominent problem of wind and solar curtailment; (2) the rigidity of computing load adjustment and the lack of new energy response capability; and (3) the large-scale centralized operation of computing centers puts structural pressure on urban power distribution networks, it is urgent to propose a new computing task scheduling scheme based on new energy consumption. Summary of the Invention

[0005] In view of this, this disclosure proposes a computing power task scheduling method, device and storage medium based on new energy consumption.

[0006] According to one aspect of this disclosure, a computing power task scheduling method based on renewable energy consumption is provided. The method includes:

[0007] Based on the historical output power data of at least one new energy power generation node and other load data that draws power from the new energy power generation node, a surplus power output prediction curve is established. The surplus power output prediction curve is used to indicate the remaining output power data that each new energy power generation node can be dispatched within a future preset time period.

[0008] Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained.

[0009] Based on the task scheduling strategy, control the computing resources to execute at least one computing task.

[0010] In one possible implementation, a surplus power output prediction curve is established based on historical output power data of at least one renewable energy generation node and other load data from which electricity is drawn on the renewable energy generation node, including:

[0011] Historical remaining output power data is obtained by comparing historical output power data with other load data.

[0012] Historical residual output power data and environmental factor data are input into the prediction model to output a residual power output prediction curve.

[0013] In one possible implementation, the method further includes:

[0014] Based on the task description information of at least one computing power task, determine the multi-dimensional feature vector corresponding to each computing power task;

[0015] Based on the multi-dimensional feature vectors corresponding to each computing power task, at least one computing power task is classified into a highly flexible task or a low-flexibility task. The schedulability of highly flexible tasks is higher than that of low-flexibility tasks.

[0016] Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained, including:

[0017] Task scheduling strategy for fixed, low-flexibility tasks;

[0018] Based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained.

[0019] In one possible implementation, based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained, including:

[0020] The surplus power output prediction curve, the multi-dimensional feature vectors corresponding to each computing power task, the computing power resource condition information, and the task scheduling strategy of low-flexibility tasks are set as scheduling constraints.

[0021] Based on scheduling constraints and a preset objective function, a task scheduling strategy for at least one computing power task is obtained.

[0022] The task scheduling strategy includes the identifier of each computing task, the start and stop times of the computing task, and the computing resource information used to process the computing task.

[0023] In one possible implementation, the multidimensional feature vector includes feature values ​​for a power consumption dimension and a flexibility dimension, where the power consumption dimension includes one or more of the following:

[0024] Average power consumption, peak power consumption, relationship between task running frequency and power consumption, GPU dependency information, and CPU dependency information during the execution of computing tasks;

[0025] The flexibility dimension includes one or more of the following:

[0026] The decomposability, parallelism, interruptibility, and latest completion time of computing tasks.

[0027] In one possible implementation, the objective function is optimized based on at least one of the following: renewable energy utilization rate, carbon emissions, and electricity cost.

[0028] In one possible implementation, based on a task scheduling strategy, computing resources are controlled to execute at least one computing task, including:

[0029] Based on the task scheduling strategy, scheduling instructions are generated, and at least one computing task is sent to the computing node indicated in the corresponding computing resource information to execute each computing task within the corresponding computing start and stop time.

[0030] In one possible implementation, the scheduling constraints also include energy storage system status information. The energy storage system stores the remaining output power from each new energy generation node and discharges it when needed for task scheduling. The task scheduling strategy also includes the charging and discharging information of the energy storage system.

[0031] The method also includes:

[0032] The energy storage system is controlled to perform charging or discharging operations within the time period indicated in the charging and discharging information.

[0033] According to another aspect of this disclosure, a computing power task scheduling device based on renewable energy consumption is provided. The device includes:

[0034] A module is established to build a surplus power output prediction curve based on the historical output power data of at least one new energy power generation node and other load data that draws power from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data of each new energy power generation node that can be dispatched within a future preset time period.

[0035] The first determining module is used to obtain the task scheduling strategy for at least one computing power task based on the surplus power output prediction curve.

[0036] The control module is used to control computing resources to execute at least one computing task based on the task scheduling strategy.

[0037] In one possible implementation, a module is created for:

[0038] Historical remaining output power data is obtained by comparing historical output power data with other load data.

[0039] Historical residual output power data and environmental factor data are input into the prediction model to output a residual power output prediction curve.

[0040] In one possible implementation, the device further includes:

[0041] The second determining module is used to determine the multi-dimensional feature vector corresponding to each computing power task based on the task description information of at least one computing power task.

[0042] The classification module is used to classify at least one computing task into a high-flexibility task or a low-flexibility task based on the multi-dimensional feature vectors corresponding to each computing task. The schedulability of high-flexibility tasks is higher than that of low-flexibility tasks.

[0043] The first determining module is used for:

[0044] Task scheduling strategy for fixed, low-flexibility tasks;

[0045] Based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained.

[0046] In one possible implementation, based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained, including:

[0047] The surplus power output prediction curve, the multi-dimensional feature vectors corresponding to each computing power task, the computing power resource condition information, and the task scheduling strategy of low-flexibility tasks are set as scheduling constraints.

[0048] Based on scheduling constraints and a preset objective function, a task scheduling strategy for at least one computing power task is obtained.

[0049] The task scheduling strategy includes the identifier of each computing task, the start and stop times of the computing task, and the computing resource information used to process the computing task.

[0050] In one possible implementation, the multidimensional feature vector includes feature values ​​for a power consumption dimension and a flexibility dimension, where the power consumption dimension includes one or more of the following:

[0051] Average power consumption, peak power consumption, relationship between task running frequency and power consumption, GPU dependency information, and CPU dependency information during the execution of computing tasks;

[0052] The flexibility dimension includes one or more of the following:

[0053] The decomposability, parallelism, interruptibility, and latest completion time of computing tasks.

[0054] In one possible implementation, the objective function is optimized based on at least one of the following: renewable energy utilization rate, carbon emissions, and electricity cost.

[0055] In one possible implementation, the control module is used for:

[0056] Based on the task scheduling strategy, scheduling instructions are generated, and at least one computing task is sent to the computing node indicated in the corresponding computing resource information to execute each computing task within the corresponding computing start and stop time.

[0057] In one possible implementation, the scheduling constraints also include energy storage system status information. The energy storage system stores the remaining output power from each new energy generation node and discharges it when needed for task scheduling. The task scheduling strategy also includes the charging and discharging information of the energy storage system.

[0058] The device also includes:

[0059] The charge / discharge module is used to control the energy storage system to perform charging or discharging operations within the time period indicated in the charge / discharge information.

[0060] According to another aspect of this disclosure, a computing power task scheduling device based on new energy consumption is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0061] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0062] According to another aspect of this disclosure, a computer program product is provided, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0063] According to embodiments of this disclosure, a surplus power output prediction curve is established based on historical output power data of at least one new energy power generation node and other load data from which power is drawn from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data available for scheduling for each new energy power generation node within a preset future time period. Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained. Based on the task scheduling strategy, computing power resources are controlled to execute at least one computing power task. This can overcome the structural limitations of rigid load in computing power centers, and from the perspective of computing power centers, achieve flexible scheduling of computing power tasks for distributed new energy local consumption scenarios, reduce curtailment rate and grid peak-shaving pressure, significantly improve the local utilization efficiency of green energy and the energy adaptability of task scheduling, and improve the energy efficiency level and green computing capabilities of computing power center operation.

[0064] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0065] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0066] Figure 1 A schematic diagram illustrating an application scenario according to an embodiment of this disclosure is shown.

[0067] Figure 2 A flowchart is shown for a computing power task scheduling method based on renewable energy consumption according to an embodiment of the present disclosure.

[0068] Figure 3 A schematic diagram of the architecture of a computing power task scheduling system based on renewable energy consumption according to an embodiment of the present disclosure is shown.

[0069] Figure 4 A structural diagram of a computing power task scheduling device based on renewable energy consumption according to an embodiment of the present disclosure is shown.

[0070] Figure 5 This is a block diagram illustrating an apparatus 1900 for scheduling computing tasks according to an exemplary embodiment. Detailed Implementation

[0071] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0072] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0073] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0074] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0075] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0076] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0077] Renewable energy is developing rapidly worldwide, with photovoltaic (PV) and wind power becoming crucial components of the power system. Numerous distributed renewable energy systems are deployed in urban parks, industrial plants, and school rooftops, creating abundant and dispersed power sources. However, due to their inherent intermittency and instability, these systems often experience a mismatch between power generation capacity and load demand, leading to frequent issues like wind and solar power curtailment. This not only wastes usable green energy resources but also increases the grid's peak-shaving and frequency regulation burden. Especially in urban areas, the complex distribution network structure, dense loads, and limited adjustability make the integration of distributed PV systems prone to problems such as localized voltage fluctuations, reverse power flow, and protection coordination failures, further complicating renewable energy integration.

[0078] Meanwhile, with the rapid rise of emerging industries such as artificial intelligence, big data, and digital twins, computing centers, as the infrastructure supporting these technologies, have become typical representatives of new high-energy-consuming loads. Large computing centers often consume hundreds of millions of kilowatt-hours of electricity annually, and their power load is highly continuous, intensive, and rigid, making it difficult to flexibly adjust according to external power supply conditions. This operational characteristic means that computing centers have almost no ability to follow the fluctuations in renewable energy output, failing to provide effective support for the local consumption of renewable energy. Furthermore, the continuous high-load operation of computing centers during peak hours further exacerbates the peak pressure on the regional power distribution network, increasing the risk of grid failures and power supply bottlenecks.

[0079] Therefore, in response to the three major practical dilemmas faced in the current scenario: (1) the mismatch between new energy power generation and local consumption capacity, and the prominent problem of wind and solar curtailment; (2) the rigidity of computing load adjustment and the lack of new energy response capability; and (3) the large-scale centralized operation of computing centers puts structural pressure on urban power distribution networks, it is urgent to propose a new computing task scheduling scheme based on new energy consumption.

[0080] Currently, most computing centers still prioritize performance, minimize latency, and maximize resource utilization in task scheduling. Their scheduling systems generally lack the ability to perceive and interact with external energy systems. This "black box" internal scheduling mechanism prevents computing centers from achieving coordinated optimization between tasks and power supply, hindering the large-scale consumption of clean energy in computing scenarios.

[0081] Existing technologies include several green energy supply or power regulation solutions for computing power centers. For example, the State Grid Qinghai Electric Power Company proposed a "Method for Constructing a Green Power Computing Power Fusion Scheduling Model and Scheduling Platform" (CN119376928A). This method constructs a multi-step scheduling approach, including "computing power center load analysis, peak shaving and valley filling to optimize load, clean energy output load prediction, optimal computing power load curve generation, and task response monitoring." The platform also possesses electricity price economic analysis and carbon emission reduction assessment functions. The core of this solution lies in optimizing the distribution of computing power load based on big data and electricity price time-series information, achieving proactive matching between computing power tasks and clean energy output, and emphasizing provincial-level scheduling, control, and analysis capabilities.

[0082] In addition, the "Method and Device for Power Supply Capacity Regulation of Computing Infrastructure" (CN118863414A) proposed by the State Grid Energy Research Institute Co., Ltd. provides a power regulation method based on a gap compensation mechanism. This scheme predicts power demand to determine whether the current local power supply capacity is sufficient. If a power gap exists, it prioritizes the use of locally adjustable equipment (such as air conditioners, energy storage, diesel generators, etc.) for start-up and shutdown. If the demand cannot be met, the load is transferred to other regions with the highest collaborative assessment scores. Its technical focus is on improving the power supply flexibility of computing centers under power shortages and constructing a cross-regional power resource coordination and dispatch system.

[0083] The embodiments disclosed herein address the following shortcomings of the prior art:

[0084] (1) The current solutions are generally dominated by the power supply side and lack the adjustment mechanism of the computing center side. Existing solutions mainly focus on the load adjustment or energy distribution logic of the power grid side, which makes it difficult to implement the scheduling strategy at the task level.

[0085] (2) The dynamic response relationship between computing power tasks and renewable energy output has not been established. The current scheme still mainly uses fixed scheduling for computing power task scheduling, and lacks feedforward utilization of green power supply capacity.

[0086] (3) It is impossible to break through the structural limitation of rigid load in computing centers. Existing grid-side scheduling mostly alleviates load pressure through external resource scheduling (such as starting and stopping equipment, regional migration), rather than starting from the computing task itself to build a flexible and adjustable operating mechanism, which makes it difficult to effectively release the flexible scheduling potential of the computing system.

[0087] (4) The optimization is not geared towards typical scenarios of local consumption of distributed renewable energy. Most solutions still rely on centralized power sources or emergency power guarantees, and lack structural coupling with factors such as the location and output characteristics of distributed renewable energy in the city, making it difficult to achieve local closed-loop utilization of regional green energy.

[0088] In view of this, this disclosure proposes a computing power task scheduling method, apparatus, and storage medium based on renewable energy consumption. The method of this disclosure establishes a surplus power output prediction curve based on historical output power data of at least one renewable energy power generation node and other load data from which power is drawn. The surplus power output prediction curve indicates the remaining output power available for scheduling at each renewable energy power generation node within a preset future time period. Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained. Based on the task scheduling strategy, computing resources are controlled to execute at least one computing power task. This method can overcome the structural limitations of rigid load on computing centers, achieving flexible scheduling of computing power tasks for distributed renewable energy local consumption scenarios from the perspective of the computing center side. This reduces the curtailment rate and grid peak-shaving pressure, significantly improves the local utilization efficiency of green energy and the energy adaptability of task scheduling, and enhances the energy efficiency and green computing capabilities of the computing center.

[0089] Figure 1 The diagram illustrates an application scenario according to an embodiment of this disclosure. The computing power task scheduling system based on renewable energy consumption, as described in this disclosure, can be applied to scenarios such as intelligent computing centers, urban edge computing nodes, and data center clusters that execute computing power tasks, thus building a new operating mechanism for computing power infrastructure that features "visible energy consumption, flexible tasks, and intelligent scheduling." Figure 1 As shown, this computing power task scheduling system based on renewable energy consumption can collect output power data of distributed renewable energy nodes and other load data that draws power from renewable energy generation nodes, and obtain computing power tasks to be executed. Based on the method of this embodiment, the computing power tasks are scheduled, and the tasks are distributed to the computing nodes of the computing center for execution according to the scheduling results. In the application scenarios of this embodiment, it can promote the efficient local consumption of distributed renewable energy, accurately align the time-of-use output of distributed renewable energy through computing power load, realize the "local and local use" of green electricity, reduce the curtailment rate and grid peak-shaving pressure, and improve the investment efficiency of renewable energy.

[0090] The computing power task scheduling system based on new energy consumption in this disclosure can be deployed on terminal devices or servers. The terminal devices involved in this disclosure can be any one or more of the following: mobile phones, foldable electronic devices, tablet computers, desktop computers, laptop computers, handheld computers, laptops, ultra-mobile personal computers (UMPCs), netbooks, cellular phones, personal digital assistants (PDAs), and in-vehicle devices. This disclosure does not impose any special limitations on the specific type of terminal device; it can have wired or wireless communication capabilities.

[0091] The server disclosed in this embodiment can be located locally or in the cloud, and can be a physical device or a virtual device, such as a virtual machine or container. It has wireless communication capabilities, which can be configured in the server's chip (system) or other components. The wireless communication capabilities can be implemented through mobile communication technologies such as 2G / 3G / 4G / 5G, as well as Wi-Fi, Bluetooth, frequency modulation (FM), data radio, and satellite communication. It can also communicate via a wired connection to enable interaction with other devices.

[0092] Figure 2 A flowchart illustrating a computing power task scheduling method based on renewable energy consumption according to an embodiment of this disclosure is shown. Figure 2 As shown, the method may include:

[0093] Step S201: Based on the historical output power data of at least one new energy power generation node and other load data that draws power from the new energy power generation node, establish a surplus power output prediction curve.

[0094] New energy sources can include any one or more clean energy sources, such as photovoltaic power generation, wind power generation, and biomass energy. New energy power generation nodes can refer to distributed clean energy devices with power generation capabilities within the area where the computing center is located. For example, they can be distributed photovoltaic modules (with a total installed capacity of, for example, 3MW) deployed on the rooftops of buildings in that area.

[0095] A computing power center can be a computing service infrastructure deployed in places such as industrial parks, office buildings or data centers. A computing power center can include multiple graphics processing unit (GPU) computing nodes and a central processing unit (CPU) resource pool as computing power resources. The computing power center can be equipped with a task scheduling platform. The task scheduling platform can use container orchestration systems such as Kubernetes to achieve unified resource scheduling and has functions such as task queue management and task execution frequency control.

[0096] The output power data of each new energy power generation node can be collected in real time by IoT data acquisition devices deployed in the area (such as Internet of Things (IoT) sensors deployed on building rooftops and power distribution cabinets) and uploaded to the aforementioned computing power task scheduling system based on new energy consumption.

[0097] It can also collect other load data that draws power from new energy power generation nodes in real time and upload them to the aforementioned computing power task scheduling system based on new energy consumption. Other load data may include non-computing power related power consumption, such as the power consumption of infrastructure such as air conditioning, lighting, and network equipment.

[0098] The collected data can also be uploaded to the database for storage and archiving at preset time intervals (such as 5 minutes).

[0099] Output power data within the most recent preset time period (e.g., within 24 hours) can be used as historical output power data. This data, combined with other load data from the same preset time period, can be used to predict the surplus power output curve. The surplus power output curve can indicate the remaining output power available for dispatch at each renewable energy generation node within a future preset time period. This future preset time period can be set according to actual needs, such as any time scale from 1 hour to 24 hours, supporting accuracy at the hourly or minute level.

[0100] In step S201, the following can be done:

[0101] Historical remaining output power data is obtained by comparing historical output power data with other load data; historical remaining output power data and environmental factor data are input into the prediction model to output the remaining power output prediction curve.

[0102] Historical surplus output power represents the surplus output power available for dispatch at each renewable energy generation node within the most recent preset time period. Environmental factor data can include meteorological information affecting the power generation capacity of renewable energy generation nodes, such as solar irradiance, temperature, and cloud cover within the next preset time period. By using historical surplus output power data and environmental factor data as joint inputs for modeling and prediction, the accuracy of surplus power output prediction can be improved.

[0103] The prediction model can be an integration of XGBoost (eXtreme Gradient Boosting), Long Short-Term Memory (LSTM) networks, Transformer, etc. It can also be implemented using simulation tools such as PVLib or related prediction methods such as formula derivation.

[0104] A prediction model can be trained based on a pre-built training set. For example, the training set can include historical (e.g., data from the most recent 24 hours ago) data on the output power of renewable energy nodes and corresponding environmental factor data for the corresponding time period, with the training label being the actual remaining output power at the corresponding moment. During training, a supervised learning approach can be used. The training data is input into the initial prediction model, which outputs a corresponding surplus power prediction curve. Then, the parameters of the initial prediction model can be optimized by minimizing the error (e.g., mean square error) between the predicted remaining output power at each moment in the surplus power prediction curve and the training label, resulting in the trained prediction model.

[0105] Step S202: Based on the surplus power output prediction curve, obtain the task scheduling strategy for at least one computing power task.

[0106] Computing tasks can represent various computationally intensive tasks waiting to run on computing nodes, such as model training tasks, inference tasks, data preprocessing tasks, image rendering tasks, etc.

[0107] The task scheduling strategy can be represented as a three-dimensional task scheduling matrix, including the identifier of each computing task, the start and stop times of the computing task, and the computing resource information used to process the computing task. The start and stop times indicate the start and stop times of the computing task. For some computing tasks, they can be split into multiple sub-tasks to be executed in different time periods. The computing resource information can include the running frequency of the computing task and the computing resource configuration on which the task depends, such as the GPU computing node executing the computing task, the required GPU memory capacity, the number of CPU cores used or the size of the CPU resource pool to which it belongs, and the memory capacity required for task execution.

[0108] First, the relevant information of computing tasks can be structured and modeled, and the computing tasks can be classified to support the formulation of subsequent load adaptation and scheduling strategies.

[0109] The method also includes:

[0110] Based on the task description information of at least one computing power task, determine the multi-dimensional feature vector corresponding to each computing power task; based on the multi-dimensional feature vector corresponding to each computing power task, classify at least one computing power task into a high-flexibility task or a low-flexibility task.

[0111] Task description information can include task type (e.g., training, inference, data preprocessing), execution time constraints (e.g., whether it's real-time, latest completion time), resource requirements (e.g., number of GPUs / CPUs, memory size, runtime), decomposability (e.g., whether the task can be divided into multiple subtasks for asynchronous execution), parallelism (e.g., whether parallel scheduling is supported within the task), and interruptibility (whether pause and resume are allowed). This task description information can be transformed into structured, numerical, multi-dimensional feature vectors by pre-constructing feature extraction rules, mapping tables, or employing natural language parsing and encoding techniques.

[0112] The multidimensional feature vector can include feature values ​​of power consumption dimension and flexibility dimension. The power consumption dimension can be used to reflect the energy consumption behavior of computing tasks, while the flexibility dimension can be used to measure the flexibility of computing tasks in scheduling.

[0113] The power consumption dimension can include one or more of the following:

[0114] The average power consumption, peak power consumption, relationship between task running frequency and power consumption during the execution of computing tasks, GPU dependency information (such as the number of GPUs required), and CPU dependency information (such as the number of CPUs required).

[0115] The flexibility dimension can include one or more of the following:

[0116] The decomposability, parallelism, interruptibility, and latest completion time of computing tasks.

[0117] Based on the aforementioned multidimensional feature vectors of computing tasks, these tasks can be processed using relevant feature engineering and classification algorithms (such as rule-based classification methods, supervised or unsupervised learning algorithms) to classify them into highly flexible or low-flexibility tasks. High-flexibility tasks have a higher degree of schedulability than low-flexibility tasks. Low-flexibility tasks typically have strict latency constraints, fixed execution patterns, and resource dependencies, resulting in limited scheduling space. In the subsequent determination of task scheduling strategies, the scheduling strategies for highly flexible tasks can be prioritized to fully unleash their scheduling potential, while information about low-flexibility tasks can be prioritized as scheduling constraints.

[0118] Through the above-mentioned task-level power consumption modeling mechanism, the differences in resource consumption, execution rigidity and time flexibility of different computing power tasks can be accurately described, which enhances the ability of subsequent scheduling to adapt to the fluctuation of new energy output.

[0119] In step S202, the following can be done:

[0120] Task scheduling strategy for fixed, low-flexibility tasks;

[0121] Based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained.

[0122] The task scheduling strategy for at least one computing power task can include a task scheduling strategy for low-flexibility tasks and a task scheduling strategy for high-flexibility tasks.

[0123] Among them, the surplus power output prediction curve, the multi-dimensional feature vectors corresponding to each computing power task, the computing power resource condition information, and the task scheduling strategy of low-flexibility tasks can be set as scheduling constraints.

[0124] Based on scheduling constraints and a preset objective function, a task scheduling strategy for at least one computing power task is obtained.

[0125] For example, based on the relevant task description information of low-flexibility tasks (such as the latest completion time, resource consumption requirements, task type, etc.), the corresponding computing start and stop times and required computing resources can be prioritized for low-flexibility tasks to obtain the task scheduling strategy for low-flexibility tasks. This information will remain unchanged as a scheduling constraint in the subsequent process of determining the scheduling strategy for all computing tasks.

[0126] In setting scheduling constraints, the surplus power output prediction curve can represent the time series constraints of new energy capacity. If the total power consumption in any scheduling time period exceeds the available power in the corresponding time period in the surplus power output prediction curve, non-new energy sources will be used for power supply.

[0127] The relevant information contained in the multi-dimensional feature vectors corresponding to each computing task (GPU dependency information, CPU dependency information, task decomposability, parallelism, interruptibility, latest completion time, etc.) can be used to constrain the power consumption and flexibility requirements in the subsequent task scheduling strategy determination process. That is, the determined task scheduling strategy needs to conform to the relevant information contained in the multi-dimensional feature vectors corresponding to each computing task. For example, for indivisible computing tasks, they cannot be split in the task scheduling strategy. Therefore, by structurally modeling computing tasks and incorporating them into scheduling constraints, computing tasks acquire adjustable response capabilities to power supply characteristics, providing a foundation for the computing power scheduling system to move towards flexible operation.

[0128] The computing resource condition information can be the current status, total capacity, and resource release time of computing resources such as GPUs / CPUs collected in real time. It can be used to constrain the resource allocation of task scheduling. That is, the determined task scheduling strategy must comply with the computing resource condition information, such as not exceeding the total capacity of computing resources.

[0129] In one possible implementation, the scheduling constraints also include energy storage system status information. The energy storage system stores the remaining output power from each renewable energy generation node and discharges it when needed for task scheduling. The energy storage system status information can represent parameters such as the current remaining capacity, maximum charge / discharge power, charge / discharge efficiency, and available charge / discharge time window of the energy storage system.

[0130] The aforementioned task scheduling strategy may also include charging and discharging information of the energy storage system. This information can represent the charging or discharging time period and power level during task execution, and is used to achieve coordinated management of the energy storage system and new energy supply while optimizing scheduling.

[0131] The objective function value can be calculated based on the task scheduling strategy, and the optimization objective of the objective function can be determined based on at least one of the following: renewable energy utilization rate, carbon emissions, and electricity cost. Single-objective optimization can be carried out based on objectives such as maximizing renewable energy utilization rate, minimizing carbon emissions, and minimizing electricity cost. For example, for the optimization objective of maximizing renewable energy utilization rate, the preset objective function can be expressed as:

[0132]

[0133] Among them, U RE For the utilization rate of new energy sources. E RE E refers to the renewable energy used in the mission. total E represents the total power required to execute the task. total It can be determined based on the cumulative power consumption of each computing task in different time periods within the task scheduling strategy. RE It could be E total The portion of the remaining output power that is available for dispatch at the corresponding moment of the predicted surplus power output curve represents the portion of the task power consumption within each time period that can be covered by renewable energy supply. Any excess power is considered to be supplied from the traditional power grid (i.e., non-clean energy) and is not included in the renewable energy utilization rate.

[0134] This can increase the proportion of green electricity used, and improve the green rating and policy competitiveness of computing centers.

[0135] For the optimization objective of minimizing carbon emissions, the pre-defined objective function can be expressed as:

[0136]

[0137] Among them, C carbon E represents total carbon emissions. t The power consumption of the computing center at time t can be determined by summing the power consumption of each computing task at time t in the task scheduling strategy, φ. t The carbon emission factor is the amount of carbon emissions per unit of electricity used at that moment. It can be determined based on the power grid structure (such as the proportion of coal power, wind power, and photovoltaic power) during that period.

[0138] For the optimization objective of minimizing electricity costs, the pre-defined objective function can be expressed as:

[0139]

[0140] Among them, C electricity p represents the total electricity cost. t Let t be the electricity price at time t.

[0141] Therefore, by using a dispatching model dominated by green energy, high-energy-consuming tasks can be concentrated in periods with low electricity prices and high green electricity availability, thereby reducing operating electricity costs.

[0142] Multi-objective optimization can also be carried out through preset weight combinations to help computing centers reduce costs, increase efficiency, and achieve green transformation. The preset objective function can then be expressed as:

[0143]

[0144] Where f is the objective function value in multi-objective optimization, and the goal is to minimize f. λ1, λ2, and λ3 are the weights of each objective in multi-objective optimization, satisfying λ1 + λ2 + λ3 = 1.

[0145] Therefore, while ensuring computing power performance, the local consumption rate of distributed photovoltaic power generation can be improved, and carbon emissions and operating electricity costs can be reduced.

[0146] A scheduling optimization problem can be constructed based on scheduling constraints and a preset objective function, and then solved using relevant optimization algorithms (such as heuristic algorithms). During the optimization process, the task scheduling strategy can be continuously adjusted to make its corresponding objective function value approach the optimal solution, thereby achieving strategy optimization and obtaining a task scheduling strategy that meets the optimization objective, which can be used to schedule various computing power tasks.

[0147] Furthermore, a rolling time-domain optimization mechanism can be introduced to reschedule and optimize the remaining unfinished computing tasks at set time intervals (such as every hour), thereby dynamically adapting to real-time changes in computing resource status, energy supply, etc., and improving the flexibility and practicality of scheduling strategies.

[0148] Therefore, this embodiment of the present disclosure uses a task scheduling optimization mechanism with surplus power output prediction curve as the leading input, and combines the multi-dimensional feature vector of the task to dynamically generate task scheduling strategies through rolling optimization or reinforcement learning. This can achieve intelligent alignment between the computing power consumption curve and the output curve of green electricity such as photovoltaic power. In this way, the new energy output prediction result (i.e., surplus power output prediction curve) is incorporated into the computing power task scheduling logic, so that the green electricity and task execution sequence are consistent, and the coordinated scheduling of "source-load" is realized.

[0149] Step S203: Based on the task scheduling strategy, control the computing resources to execute at least one computing task.

[0150] According to embodiments of this disclosure, a surplus power output prediction curve is established based on historical output power data of at least one new energy power generation node and other load data from which power is drawn from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data available for scheduling for each new energy power generation node within a preset future time period. Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained. Based on the task scheduling strategy, computing power resources are controlled to execute at least one computing power task. This can overcome the structural limitations of rigid load in computing power centers, and from the perspective of computing power centers, achieve flexible scheduling of computing power tasks for distributed new energy local consumption scenarios, reduce curtailment rate and grid peak-shaving pressure, significantly improve the local utilization efficiency of green energy and the energy adaptability of task scheduling, and improve the energy efficiency level and green computing capabilities of computing power center operation.

[0151] In step S203, the following can be done:

[0152] Based on the task scheduling strategy, scheduling instructions are generated, and at least one computing task is sent to the computing node indicated in the corresponding computing resource information to execute each computing task within the corresponding computing start and stop time.

[0153] For example, task scheduling policies can be converted into scheduling instructions that can be recognized by computing task orchestration and scheduling platforms such as Kubernetes, ARN, and Slurm (e.g., in a Kubernetes scenario, they can be converted into Pod scheduling instructions). Scheduling instructions can be sent to the corresponding compute nodes through the scheduling control interface to control operations such as starting, pausing, rescheduling, and adjusting the running frequency of computing tasks, thereby achieving dynamic scheduling control and energy consumption management of actual task execution.

[0154] During the execution of computing tasks, information such as actual power consumption, completion progress, and operating status of computing nodes can be continuously collected and compared with the predicted expected data. When the deviation between the monitored actual data and the predicted expected data exceeds a preset threshold, the rolling correction logic of the task scheduling strategy can be triggered to automatically optimize and adjust the remaining scheduling plan to ensure that the overall scheduling goal is not significantly damaged.

[0155] For example, when a highly flexible task is detected to have completed ahead of schedule, the relevant computing resources can be automatically released, and subsequent tasks can be moved forward. The optimization process in step S202 can be re-executed to generate a new round of task scheduling strategy.

[0156] For example, when the real-time power consumption of a highly flexible task exceeds the available output power value indicated by the surplus power output prediction curve and exceeds a preset threshold, the operating frequency of the task can be automatically reduced to avoid power shortage.

[0157] In one possible implementation, the method may further include:

[0158] The energy storage system is controlled to perform charging or discharging operations within the time period indicated in the charging and discharging information.

[0159] Therefore, when the fluctuation between the predicted surplus power output curve and the actual output power of renewable energy nodes exceeds a set threshold, or when there is a significant gap between the power consumption requirements of computing tasks and the output power of renewable energy nodes, the energy storage system can smooth the energy curve, ensuring the stability and sustainability of critical task operation. For example, when there is surplus power, the energy storage system can enter a charging state; while when renewable energy output decreases in the afternoon but the task load is still at its peak, the energy storage system can draw on electricity to discharge, providing supplementary power for computing tasks, reducing the risk of task interruption and decreasing dependence on traditional energy sources.

[0160] In this embodiment, after each round of scheduling, the carbon reduction effect and economic benefits of the computing power center can be quantified based on multi-dimensional indicators such as the actual proportion of renewable energy use, carbon emission factor, time-of-use electricity price curve, renewable energy subsidies, and participation response rewards for each computing power task in each time period. A carbon footprint and economic analysis report (e.g., a carbon emission report and economic analysis chart) can be generated for use in green computing power certification or carbon asset assessment, providing data support for achieving green computing power certification and carbon asset trading. Furthermore, the proportion of renewable energy use during the operation of computing power tasks can be recorded, carbon emission reductions calculated, and uploaded to the carbon footprint analysis platform.

[0161] Figure 3 A schematic diagram of the architecture of a computing power task scheduling system based on renewable energy consumption, according to an embodiment of this disclosure, is shown. Figure 3 As shown, the computing power task scheduling system based on new energy consumption in this embodiment of the present disclosure may include a distributed new energy acquisition and prediction module, a computing power task modeling module, a task scheduling optimization module, a scheduling execution control module, an energy storage system coordination module, and a carbon emission and economic analysis module.

[0162] The distributed new energy acquisition and prediction module can monitor and predict the future output of local distributed photovoltaic and other green electricity in real time, and input the results (i.e., the surplus power output prediction curve mentioned above) into the task scheduling optimization module. This module, together with the task energy consumption flexibility information (i.e., the multi-dimensional feature vector mentioned above) provided by the computing power task modeling module, generates the optimal scheduling strategy (i.e., the task scheduling strategy of at least one computing power task mentioned above). This strategy is then distributed to each computing node in the computing power center by the scheduling execution control module to complete the actual task deployment. At the same time, the energy storage system coordination module (optionally) can be used to support energy regulation (i.e., control the charging and discharging of the energy storage system), and the carbon emission and economic analysis module can support the quantification of carbon reduction effects and economic benefits. Unlike existing technologies where computing power task scheduling systems lack a unified interface and feedback path with energy-side control, making it difficult to optimize task execution driven by new energy output in actual operation, this disclosed embodiment constructs a collaborative system architecture including the aforementioned modules and interfaces with mainstream computing power task scheduling platforms. This enables closed-loop collaborative control from green energy information perception to task scheduling decision-making, and then to task resource allocation and energy consumption control. It constructs a full-link closed-loop control system of "prediction-modeling-optimization-control-feedback", forming a complete end-to-end dynamic scheduling control path. It has good technical closedness and engineering deployability, and can be widely applied to heterogeneous computing power centers and different new energy combination scenarios.

[0163] This disclosure uses renewable energy output prediction as a guide, integrating computing power task energy consumption modeling and flexible task scheduling logic, and for the first time realizes a system-level operation mode of "computing following power generation, computing and power interaction, and computing and power collaboration". Compared with the traditional method of "rigid task execution and passive power supply from the grid", this disclosure realizes active matching and scheduling between renewable energy output and computing power consumption, significantly improving the local utilization efficiency of green energy and the energy adaptability of task scheduling.

[0164] Figure 4 A structural diagram of a computing power task scheduling device based on renewable energy consumption according to an embodiment of this disclosure is shown. Figure 4 As shown, the device includes:

[0165] Module 401 is used to establish a surplus power output prediction curve based on the historical output power data of at least one new energy power generation node and other load data that draws power from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data of each new energy power generation node that can be dispatched within a future preset time period.

[0166] The first determining module 402 is used to obtain a task scheduling strategy for at least one computing power task based on the surplus power output prediction curve.

[0167] The control module 403 is used to control computing resources to execute at least one computing task based on a task scheduling strategy.

[0168] In one possible implementation, module 401 is established for:

[0169] Historical remaining output power data is obtained by comparing historical output power data with other load data.

[0170] Historical residual output power data and environmental factor data are input into the prediction model to output a residual power output prediction curve.

[0171] In one possible implementation, the device further includes:

[0172] The second determining module is used to determine the multi-dimensional feature vector corresponding to each computing power task based on the task description information of at least one computing power task.

[0173] The classification module is used to classify at least one computing task into a high-flexibility task or a low-flexibility task based on the multi-dimensional feature vectors corresponding to each computing task. The schedulability of high-flexibility tasks is higher than that of low-flexibility tasks.

[0174] The first determining module 402 is used for:

[0175] Task scheduling strategy for fixed, low-flexibility tasks;

[0176] Based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained.

[0177] In one possible implementation, based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, a task scheduling strategy for at least one computing power task is obtained, including:

[0178] The surplus power output prediction curve, the multi-dimensional feature vectors corresponding to each computing power task, the computing power resource condition information, and the task scheduling strategy of low-flexibility tasks are set as scheduling constraints.

[0179] Based on scheduling constraints and a preset objective function, a task scheduling strategy for at least one computing power task is obtained.

[0180] The task scheduling strategy includes the identifier of each computing task, the start and stop times of the computing task, and the computing resource information used to process the computing task.

[0181] In one possible implementation, the multidimensional feature vector includes feature values ​​for a power consumption dimension and a flexibility dimension, where the power consumption dimension includes one or more of the following:

[0182] Average power consumption, peak power consumption, relationship between task running frequency and power consumption, GPU dependency information, and CPU dependency information during the execution of computing tasks;

[0183] The flexibility dimension includes one or more of the following:

[0184] The decomposability, parallelism, interruptibility, and latest completion time of computing tasks.

[0185] In one possible implementation, the objective function is optimized based on at least one of the following: renewable energy utilization rate, carbon emissions, and electricity cost.

[0186] In one possible implementation, the control module 403 is used for:

[0187] Based on the task scheduling strategy, scheduling instructions are generated, and at least one computing task is sent to the computing node indicated in the corresponding computing resource information to execute each computing task within the corresponding computing start and stop time.

[0188] In one possible implementation, the scheduling constraints also include energy storage system status information. The energy storage system stores the remaining output power from each new energy generation node and discharges it when needed for task scheduling. The task scheduling strategy also includes the charging and discharging information of the energy storage system.

[0189] The device also includes:

[0190] The charge / discharge module is used to control the energy storage system to perform charging or discharging operations within the time period indicated in the charge / discharge information.

[0191] According to embodiments of this disclosure, a surplus power output prediction curve is established based on historical output power data of at least one new energy power generation node and other load data from which power is drawn from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data available for scheduling for each new energy power generation node within a preset future time period. Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained. Based on the task scheduling strategy, computing power resources are controlled to execute at least one computing power task. This can overcome the structural limitations of rigid load in computing power centers, and from the perspective of computing power centers, achieve flexible scheduling of computing power tasks for distributed new energy local consumption scenarios, reduce curtailment rate and grid peak-shaving pressure, significantly improve the local utilization efficiency of green energy and the energy adaptability of task scheduling, and improve the energy efficiency level and green computing capabilities of computing power center operation.

[0192] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0193] This disclosure also provides a computing power task scheduling device based on new energy consumption, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0194] This disclosure also provides a non-volatile computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0195] This disclosure also provides a computer program product, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above method.

[0196] Figure 5 This is a block diagram illustrating an apparatus 1900 for computing task scheduling according to an exemplary embodiment. For example, apparatus 1900 may be provided as a server or terminal device. (Refer to...) Figure 5 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0197] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0198] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0199] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0200] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0201] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer-readable program instructions. These electronic circuits can execute computer-readable program instructions to implement various aspects of this disclosure.

[0202] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should 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-readable program instructions.

[0203] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0204] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0205] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0206] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A computing power task scheduling method based on renewable energy consumption, characterized in that, The method includes: Based on the historical output power data of at least one new energy power generation node and other load data from which the new energy power generation node draws power, a surplus power output prediction curve is established. The surplus power output prediction curve is used to indicate the remaining output power data of each new energy power generation node that can be dispatched within a future preset time period. Based on the surplus power output prediction curve, a task scheduling strategy for at least one computing power task is obtained. Based on the task scheduling strategy, control the computing resources to execute the at least one computing task.

2. The method according to claim 1, characterized in that, The step of establishing a surplus power output prediction curve based on historical output power data of at least one new energy power generation node and other load data from which power is drawn from the new energy power generation node includes: Historical remaining output power data is obtained based on the difference between the historical output power data and the other load data. The historical remaining output power data and environmental factor data are input into the prediction model, and the remaining power output prediction curve is output.

3. The method according to claim 1, characterized in that, The method further includes: Based on the task description information of the at least one computing power task, determine the multi-dimensional feature vector corresponding to each computing power task; Based on the multi-dimensional feature vectors corresponding to each computing power task, the at least one computing power task is classified into a highly flexible task or a low-flexibility task, wherein the schedulability of the highly flexible task is higher than that of the low-flexibility task. The task scheduling strategy for at least one computing power task based on the surplus power output prediction curve includes: The task scheduling strategy for the low-flexibility task is fixed; Based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, the task scheduling strategy for the at least one computing power task is obtained.

4. The method according to claim 3, characterized in that, The task scheduling strategy for the at least one computing power task is obtained based on the surplus power output prediction curve and the multi-dimensional feature vectors corresponding to each computing power task, including: The surplus power output prediction curve, the multi-dimensional feature vectors corresponding to each computing task, the computing resource condition information, and the task scheduling strategy of the low-flexibility task are set as scheduling constraints. Based on the scheduling constraints and the preset objective function, the task scheduling strategy for the at least one computing power task is obtained; The task scheduling strategy includes the identifier of each computing task, the start and stop times of the computing task, and the computing resource information used to process the computing task.

5. The method according to claim 3 or 4, characterized in that, The multidimensional feature vector includes feature values ​​of a power consumption dimension and a flexibility dimension, wherein the power consumption dimension includes one or more of the following: Average power consumption, peak power consumption, relationship between task running frequency and power consumption, GPU dependency information, and CPU dependency information during the execution of computing tasks; The flexible dimension includes one or more of the following: The decomposability, parallelism, interruptibility, and latest completion time of computing tasks.

6. The method according to claim 4, characterized in that, The optimization objective of the objective function is determined based on at least one of the following: new energy utilization rate, carbon emissions, and electricity cost.

7. The method according to claim 4, characterized in that, The step of controlling computing resources to execute the at least one computing task based on the task scheduling strategy includes: Based on the task scheduling strategy, a scheduling instruction is generated, and the at least one computing task is sent to the computing node indicated in the corresponding computing resource information to execute each computing task within the corresponding computing start and stop time.

8. The method according to claim 6, characterized in that, The scheduling constraints also include energy storage system status information. The energy storage system stores the remaining output power from each new energy power generation node and discharges it when needed for task scheduling. The task scheduling strategy also includes the charging and discharging information of the energy storage system. The method further includes: The energy storage system is controlled to perform charging or discharging operations within the time period indicated in the charging and discharging information.

9. A computing power task scheduling device based on new energy consumption, characterized in that, The device includes: A module is established to build a surplus power output prediction curve based on the historical output power data of at least one new energy power generation node and other load data that draws power from the new energy power generation node. The surplus power output prediction curve is used to indicate the remaining output power data of each new energy power generation node that can be dispatched within a future preset time period. The first determining module is used to obtain a task scheduling strategy for at least one computing power task based on the surplus power output prediction curve. The control module is used to control the computing resources to execute the at least one computing task based on the task scheduling strategy.

10. A computing power task scheduling device based on renewable energy consumption, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

11. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

12. A computer program product comprising a computer program, or a non-volatile computer-readable storage medium carrying a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images