Data center computing power and electric power collaborative optimization scheduling method and system

Abstract

The invention discloses a computing power and electric power collaborative optimization scheduling method and system for a data center. The method comprises the following steps: based on calculation task attributes borne by a data center, distinguishing delay-sensitive and delay-insensitive workloads, associating energy consumption characteristics of a server, operation rules of energy storage equipment and power supply use efficiency of the data center, and constructing a calculation power model; establishing a stable output characteristic rule of an adjustable unit and a probabilistic output rule of an unadjustable unit based on the type of the generator set, and constructing a power model in combination with load resource parameters and adjustment characteristics; an iterative double-layer model, an upper-layer power model and a lower-layer computing power model are constructed as a framework, iterative optimization of power and computing power elements is completed based on the iterative double-layer model, and a data center computing power and power bidirectional collaborative optimization scheduling strategy is obtained for a computing power node and a power node to execute scheduling operation. According to the method, efficient collaboration and optimization between the electric power and the computing power of the data center are promoted.

Description

Technical Field

[0001] This invention relates to a method for coordinating computing power and power, and more particularly to a method and system for optimizing the scheduling of computing power and power in data centers, belonging to the field of power-computing power coordination in the construction of green data centers. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence, big data, and cloud computing, computing demand has exploded, driving the rapid expansion of computing infrastructure. However, the continuous expansion of computing resources has also brought high energy costs and enormous carbon emission pressure. At the same time, high-density computing clusters have caused significant load impacts on local power grids, affecting the stability and security of grid operation. How to coordinate the contradiction between computing development and energy consumption has become a critical issue that urgently needs to be addressed. Summary of the Invention

[0003] Purpose of the invention: To address the above-mentioned problems, this invention provides a data center computing power and power collaborative optimization scheduling method and system. By constructing a two-way collaborative mechanism between computing power and power, it achieves comprehensive improvement in data center energy consumption cost optimization and energy utilization efficiency while ensuring the quality of computing power task services and the stable operation of the power system.

[0004] Technical solution: According to a first aspect of the present invention, a data center computing power and power collaborative optimization scheduling method and system includes the following steps:

[0005] Based on the computing task attributes carried by the data center, latency-sensitive and latency-insensitive workloads are distinguished. Time transfer constraints for delayable workloads and spatial migration constraints for real-time workloads are established respectively. Based on the time transfer constraints for delayable workloads, the spatial migration constraints for real-time workloads, and the spatiotemporal distribution of workloads, the total workload of the data center in each time unit is determined. According to the total workload, server energy consumption characteristics, energy storage equipment operation rules, and data center power usage efficiency, a computing power model is constructed. The computing power model indicates the power consumption calculation rules corresponding to the current computing power demand.

[0006] Based on the generator set type, stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets are established respectively. Based on the stable output characteristic rules for adjustable generator sets, the probabilistic output rules for non-adjustable generator sets, load resource parameters and regulation characteristics, a power model is constructed. The power model indicates the total output calculation rules and load regulation constraint rules of the generator unit.

[0007] Based on the power consumption calculation rules, the total output calculation rules of the power generation unit, and the load regulation constraint rules, an iterative two-layer model is constructed. The iterative two-layer model is based on the upper-layer power model optimization and the lower-layer computing power model optimization.

[0008] Based on an iterative two-layer model, a closed-loop process is completed to iteratively optimize power and computing power elements by generating power consumption curves, performing power model optimization based on power consumption curves, generating adjustment instructions, performing computing power model optimization based on adjustment instructions, and updating power consumption curves. This yields a bidirectional collaborative optimization scheduling strategy for data center computing power and power, which is then used by computing power nodes and power nodes to perform scheduling operations.

[0009] Furthermore, the time transfer constraints of the deferred workload include: all workloads of the deferred workload must be completed before a preset deadline, and at any single moment, the amount of unprocessed deferred workload does not exceed the initial total workload; and the workload is allocated to different times for execution according to scheduling requirements within the time interval from task receipt to deadline.

[0010] Furthermore, the spatial migration constraints of the real-time workload include: the real-time workload must maintain continuous operation without interruption when migrating between different nodes; the target node must have hardware configuration compatible with the source node and sufficient free memory capacity; and the amount of data migrated in a single migration must not exceed the maximum transmission bandwidth limit of the network link between nodes.

[0011] Furthermore, determining the total workload of the data center in each time unit based on the time transition constraints of deferred workloads, the spatial migration constraints of real-time workloads, and the spatiotemporal distribution of workloads includes:

[0012] The time allocation of deferred workloads is determined based on the time transfer constraints of deferred workloads, and the spatial distribution of real-time workloads is determined based on the spatial migration constraints of real-time workloads. Combining the time allocation of deferred workloads and the spatial distribution of real-time workloads, the total amount of deferred workloads and real-time workloads that all nodes need to process in each time unit is integrated to obtain the total workload that the data center needs to bear in each time unit. The time unit is a preset fixed time interval, and the total workload must simultaneously match the node computing power configuration and task execution requirements in each time unit.

[0013] Furthermore, the computing power model is constructed based on the total workload, server energy consumption characteristics, energy storage device operating rules, and data center power utilization efficiency. This computing power model indicates the power consumption calculation rules corresponding to the current computing power demand, including:

[0014] Based on the total workload in each time unit, and combined with the correlation between server energy consumption, the number of active servers, and the intensity of the load, the total power consumption of the server is calculated using the following formula:

[0015]

[0016] in, Let K be the total power consumption of the k-th type of server. Let be the peak power of the k-th type of server. Let k be the idle power of the server. The workload that the data center needs to handle on the k-th type of servers is determined by the number of k-th type servers in the data center and the total workload of the data center. get, The number of k-th type servers currently in operation. Let k be the processing speed of the k-th type of server;

[0017] The calculated total power consumption of the server is corrected by using data center power efficiency to obtain the corrected total power consumption of the server.

[0018] The algebraic difference between the charging and discharging power of all energy storage devices is superimposed. The summation of the algebraic difference is added to the corrected total power consumption of the server to obtain the power consumption quantification result that matches the current computing power demand. A computing power model is then established, which indicates the power consumption calculation rules corresponding to the current computing power demand.

[0019] Furthermore, the stable output characteristic rules of the adjustable unit include: the power generation output of the adjustable unit is within its own technically permissible range, and the output variation in adjacent time periods does not exceed its own ramp-up capability limit, maintaining a stable output according to a preset plan or adjustment command.

[0020] Furthermore, the probabilistic output rules of the non-adjustable units include: the power generation output of wind turbines and photovoltaic power station units is based on the probabilistic distribution characteristics of natural energy resources and the segmented power characteristics of the units themselves. The probabilistic distribution characteristics of natural energy resources include the Weibull distribution of wind speed two parameters corresponding to wind turbines and the Beta distribution of solar radiation corresponding to photovoltaic power stations. The unit output is calculated by combining the predicted value of energy resources under the above probabilistic distribution with the segmented power characteristics of the units, and the output result shows a probabilistic fluctuation characteristic with the change of natural energy resources, which cannot be actively changed by adjustment commands.

[0021] Furthermore, the power model is constructed based on the stable output characteristic rules of adjustable units, the probabilistic output rules of non-adjustable units, load resource parameters, and regulation characteristics. This power model indicates the total output calculation rules for power generation units and the load regulation constraint rules, including:

[0022] By integrating the stable output characteristic rules of adjustable units and the probabilistic output rules of non-adjustable units, and superimposing the expected power consumption, minimum power, maximum power, and maximum adjustable power parameters of load resources, and combining the load resource adjustment types identified by the characteristic matrix, a power model is established that includes generation-side output calculation logic and load-side adjustment constraint logic. This forms the total output calculation rules for generation units and the load adjustment constraint rules. The total output calculation rules for generation units are used to determine the total output composition of a single generation unit, while the load adjustment constraint rules are used to regulate the adjustment range and adjustment method of load resources.

[0023] Furthermore, the iterative two-layer model, through a closed-loop process of generating electricity consumption curves, performing power model optimization based on electricity consumption curves, generating adjustment instructions, performing computing power model optimization based on adjustment instructions, and updating electricity consumption curves, completes the iterative optimization of power and computing power elements, including:

[0024] Based on the power consumption calculation rules, the computing power side basic data is input, including the type of computing task, total computing volume, priority, earliest start time and latest completion time. The computing tasks are pre-arranged to generate the initial computing power node's expected power consumption curve.

[0025] Based on the total output calculation rules of the power generation unit and the load regulation constraint rules, the power side basic data is input. The power side basic data includes the daily power generation plan curve, electricity price information, and adjustable resource regulation compensation information. The initial computing node's expected electricity consumption curve is used as the basic data. The upper-level power model optimization is performed using a mixed integer linear programming algorithm to generate resource regulation plans and regulation instructions.

[0026] Based on the computing power model, the adjustment instructions are invoked, and a genetic algorithm with an elite retention strategy is used to adjust the task scheduling scheme and energy storage charging and discharging plan, generating an updated power consumption curve for the computing power node, thus completing the optimization of the lower-level computing power model.

[0027] Calculate the norm of the expected power consumption curve of the computing node generated in two consecutive iterations. If the norm is less than a set threshold or the number of iterations reaches the preset maximum number of iterations, output the collaborative optimization scheduling strategy. If the convergence condition is not met, use the updated power consumption curve as the new basic data and return to the upper-level power model optimization step to repeat the iteration until the convergence condition is met.

[0028] Furthermore, the upper-level power model optimization takes minimizing the total cost of the power system as the objective function. The total cost of the power system includes the power generation cost of generator sets, the operating loss and aging cost of energy storage equipment, the adjustment load compensation cost, and the adjustment cost related to electricity prices. The lower-level computing power model optimization takes minimizing the total operating cost of the data center as the objective function. The total operating cost includes server energy consumption cost, task migration cost, service quality latency penalty cost, and energy storage equipment adjustment compensation cost.

[0029] According to a second aspect of the present invention, a data center computing power and power collaborative optimization scheduling system includes:

[0030] The computing power model construction module is used to distinguish between latency-sensitive and latency-insensitive workloads based on the computing task attributes carried by the data center, establish time transfer constraints for deferred workloads and spatial migration constraints for real-time workloads, determine the total workload of the data center in each time unit based on the time transfer constraints of deferred workloads, the spatial migration constraints of real-time workloads, and the spatiotemporal distribution of workloads, and construct a computing power model based on the total workload, server energy consumption characteristics, energy storage equipment operation rules, and data center power usage efficiency. The computing power model indicates the power consumption calculation rules corresponding to the current computing power demand.

[0031] The power model construction module is used to establish stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets based on generator set type. Based on the stable output characteristic rules for adjustable generator sets, the probabilistic output rules for non-adjustable generator sets, load resource parameters and regulation characteristics, the power model is constructed. The power model indicates the total output calculation rules and load regulation constraint rules of the power generation unit.

[0032] The joint optimization solution module is used to construct an iterative two-layer model based on power consumption calculation rules, total output calculation rules of power generation units, and load regulation constraint rules. The iterative two-layer model is structured with upper-layer power model optimization and lower-layer computing power model optimization. Based on the iterative two-layer model, a closed-loop process is completed to optimize power and computing power elements, resulting in a data center computing power and power bidirectional collaborative optimization scheduling strategy for scheduling operations by computing power nodes and power nodes.

[0033] According to a third aspect of the present invention, a computer device is provided, the device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs, when executed by the processors, implementing the steps of the data center computing power and power collaborative optimization scheduling method as described in the first aspect of the present invention.

[0034] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the data center computing power and power collaborative optimization scheduling method as described in the first aspect of the present invention.

[0035] According to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the data center computing power and power collaborative optimization scheduling method as described in the first aspect of the present invention.

[0036] Beneficial effects:

[0037] (1) This invention achieves more refined modeling and scheduling capabilities for data center computing loads by constructing a computing power model that distinguishes between latency-sensitive and latency-insensitive workloads. By establishing time transition constraints for delayable workloads and spatial migration constraints for real-time workloads, and finally synthesizing the total workload, this invention enables the optimization model to accurately identify and utilize the adjustability of tasks, providing a precise model foundation for flexible scheduling of computing resources in the time and space dimensions. This is a prerequisite for achieving collaboration with the power system.

[0038] (2) This invention improves the accuracy of the power sector's characterization of the uncertainty of renewable energy by establishing a probabilistic output model that comprehensively considers both adjustable and non-adjustable units. Traditional power dispatch often oversimplifies the handling of wind power and photovoltaic power. This invention models wind speed and solar radiation separately to obtain the expected output of wind turbines and photovoltaic power plants. This allows the upper-level power optimization model to more accurately reflect the power generation capacity and fluctuation range of non-adjustable units, thereby issuing more scientific and feasible adjustment instructions to the lower-level computing nodes.

[0039] (3) This invention constructs a data center computing power and power collaborative optimization scheduling model based on an iterative two-layer model by refining the adjustable computing power and power elements of the data center. It interacts and iteratively optimizes the two-layer elements of power and computing power in the data center, realizing bidirectional collaborative optimization of computing power and power in the data center. This solves the problems of loose model coupling, simple constraint elements, and conflicting optimization objectives in traditional single-layer or unidirectional optimization. Existing solutions are mostly unidirectional instructions or simple price responses, and the modeling of various elements is not sufficient, so deep collaboration cannot be achieved. The two-layer model of this invention uses a closed-loop iterative mechanism of "pre-arrangement of electricity consumption curves - power optimization - issuance of adjustment instructions - adjustment of computing power response - curve update" to ensure that the operation constraints on the power side and the service quality constraints on the computing power side can be satisfied simultaneously and dynamically in the iterative solution process, and finally converge to an optimal solution that takes into account both the economic operation of the power grid and the efficient service of computing power. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of data center computer collaborative adjustment according to an embodiment of the present invention;

[0041] Figure 2 This is a flowchart of a method according to an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram comparing the power computing power adjustment before and after according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the embodiments provided below are merely for the purpose of fully and completely disclosing the present invention and fully conveying the technical concept of the invention to those skilled in the art. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. The terminology used in the exemplary embodiments shown in the accompanying drawings is not intended to limit the invention.

[0044] Against the backdrop of the continuous expansion of computing infrastructure, exploring the synergistic mechanism between computing resources and the power system, especially utilizing adjustable load resources to achieve two-way interaction between computing power and electricity, has become an effective way to reduce computing energy consumption costs and improve the efficiency and stability of power grid operation. By guiding the flexible scheduling of computing loads in time and space, it is possible not only to alleviate power pressure during peak hours but also to effectively absorb renewable energy and improve energy utilization efficiency.

[0045] However, most current technologies focus on either computing power scheduling or power scheduling, with little research on combining the two for coordinated regulation. This scheduling model leads to insufficient optimization of energy consumption costs in data centers, failing to fully utilize real-time electricity prices and regulation signals from the power system, and also making it difficult to use the computing load of data centers as a flexible resource to support the stable operation of the power grid. Therefore, research on coordinated optimization schemes for computing power and power can promote the development of intelligent computing and other data centers towards green, low-carbon, and energy-saving directions. In view of this, this invention provides a method and system for coordinated optimization scheduling of computing power and power in data centers, achieving a comprehensive improvement in data center energy consumption cost optimization and energy utilization efficiency.

[0046] Figure 1 This is a schematic diagram of the collaborative scheduling of data center computing in an embodiment of the present invention. It includes two core sides and related entities, namely the computing power side (including latency-sensitive workloads, latency-insensitive workloads, server clusters, and energy storage devices in the data center), the power side (including adjustable units, non-adjustable units such as wind power / photovoltaic power, and load resources), and the upper-level scheduling center (computer equipment that executes the iterative two-layer model) responsible for collaborative scheduling. Figure 1 In this invention, the power generation facilities on the power side (also referred to as power nodes in this paper) include adjustable and non-adjustable generating units, receiving power generation plan adjustment instructions from the power management center. The computing power side includes computing power nodes and computing power loads. Computing power nodes are the computing servers, storage, energy storage, and other equipment supporting computing in the data center. Computing power loads are the computing tasks to be executed, broadly categorized as deferred tasks and non-deferred real-time tasks, receiving computing task assignment instructions from the computing power scheduling center. The power management center and the computing power scheduling center are interconnected through a power-computing power collaborative scheduling module, running an iterative optimization model to achieve dynamic matching of power supply and computing power demand in time and space, thereby improving energy efficiency, promoting the consumption of new energy sources, and ensuring the quality of computing power services. In the context of this invention, the power management center and the computing power scheduling center interconnected through the power-computing power collaborative scheduling module are collectively referred to as the upper-level scheduling center, which can use computer equipment executing the iterative two-layer model proposed in this invention to perform the corresponding functions.

[0047] Reference Figure 2 This invention provides a data center computing power and power collaborative optimization scheduling method. First, it models the characteristics of power and computing power nodes respectively, and determines the optimization objectives and constraints of power and computing power to construct an iterative two-layer model. Then, it inputs basic data and performs an iterative optimization process of pre-arranging power consumption curves, optimizing power, issuing adjustment instructions, adjusting computing power response, and updating curves to achieve the goal of bidirectional collaborative optimization scheduling of computing power and power nodes.

[0048] The method specifically includes the following steps:

[0049] Step S1, Construct a computing power model: Analyze the computing tasks carried by the data center, calculate and model the required computing power, and then obtain the power consumption corresponding to the computing power demand.

[0050] This invention distinguishes between latency-sensitive and latency-insensitive workloads based on the computing task attributes carried by data centers. It establishes time transfer constraints for delayable workloads and spatial migration constraints for real-time workloads. Combining the spatiotemporal distribution of workloads, it determines the total workload of each time unit in the data center. Then, it associates server energy consumption characteristics, energy storage device operation rules, and data center power usage efficiency to construct a computing power model and obtain power consumption calculation rules corresponding to computing power requirements.

[0051] S11: Obtain the workload model. Data center computing tasks can be divided into two categories: latency-sensitive and latency-insensitive. Latency-sensitive tasks need to be executed in real time, also known as real-time workloads, such as bank transfers and intelligent real-time question answering. Latency-insensitive tasks only need to be completed before the deadline and have a certain flexible time window, also known as deferred workloads, such as large model pre-training tasks, which can be carried out during off-peak hours at night after receiving the task.

[0052] The characteristics of deferred workloads are modeled. For deferred workloads, the core constraint is that the total workload of the tasks must be completed before the deadline, and at any given time, the number of tasks to be processed cannot exceed the initial total. The workload can be allocated to different times for execution according to scheduling requirements within the time interval from task receipt to the deadline. This invention refers to this as the time transition constraint of deferred workloads. Its mathematical model can be expressed as:

[0053]

[0054] in, The workload that can be delayed before responding, This represents the percentage of delayable workloads among all task types. For all types of workloads prior to response, Let t be the workload that can be delayed after responding at time t. Let be the time transition response quantity that can be delayed at time t. Let t be the time response transfer amount of the remaining workload at time t, where T is the total execution time of the task.

[0055] This invention models the scheduling and control of real-time workloads. When real-time workloads migrate between different nodes, their operation must remain continuous and uninterrupted; the target node must have hardware configurations compatible with the source node and sufficient free memory capacity; the amount of data migrated in a single instance must not exceed the maximum transmission bandwidth limit of the network link between nodes. This is referred to as the spatial migration constraint for real-time workloads. The specific formula is as follows:

[0056]

[0057] in, To respond to the workload of the previous real-time workload. This represents the percentage of real-time workloads among all task types. For all types of workloads prior to response, Let t represent the workload of the real-time workload after the response at time t. Spatial response transfer amount for real-time workloads d represents the current data center, and d represents the servers in the data center.

[0058] Based on the modeling of deferred and real-time workloads, and considering the spatial and temporal distribution of workloads, the workload that the data center needs to handle in each time unit can be obtained as follows:

[0059]

[0060] in, This refers to the workload that the data center needs to handle in each unit of time. The workload that can be delayed at time t, The real-time workload at time t.

[0061] S12: Calculate the computing power required by the data center servers based on the workload model.

[0062] After obtaining the data center's workload, server energy consumption can be calculated. Server energy consumption is primarily affected by the number of active servers and the intensity of the workload. Data centers have various types of servers, and the energy consumption of each type is detailed in the formula:

[0063]

[0064] in, Let K be the total energy consumption of the k-th type of server. Let be the peak power of the k-th type of server. Let k be the idle power of the server. The workload that the data center needs to handle on the k-th type of servers is determined by the number of k-th type servers in the data center and the total workload of the data center. get, The number of k-th type servers currently in operation. Let be the processing speed of the k-th type of server.

[0065] When calculating server energy consumption, two limitations should be considered: the maximum number of servers and the performance limitations of the servers themselves, as shown in the formula:

[0066]

[0067] in, The number of k-th type servers currently in operation. The number of the k-th type of servers that are shut down. For capacity coefficient, Let k be the total number of servers of type k. Let be the demand for the k-th type of server at time t. The processing speed of the k-th type of server when executing tasks. This represents the maximum voltage that the k-th type of server can obtain.

[0068] S13: To ensure the continuous and stable operation of the server, consider configuring backup power supplies and other energy storage devices at the computing nodes.

[0069] To ensure continuous power supply to servers, energy storage devices are typically installed in data centers as backup power to guarantee high availability. This invention incorporates this into the computing power model for unified consideration. Furthermore, the energy storage devices can participate in the data center's power regulation of each server. Assuming the data center is equipped with v energy storage devices, the specific definition of which is as follows:

[0070]

[0071] in, For the hth energy storage device, This represents the maximum available capacity of the h-th energy storage device. This refers to the maximum charging power of the energy storage device. This represents the maximum discharge power of the energy storage device. The charging efficiency of energy storage devices. The discharge efficiency of energy storage devices. This represents the maximum state of charge (SOC) of the energy storage device. This is the minimum state of charge factor for energy storage devices. This represents the total number of energy storage devices.

[0072] S14: Determine the electrical power required by the computing server based on the computing power required for the computing load.

[0073] Considering the energy consumption of IT equipment such as servers and other energy-consuming devices within a data center, the specific formula is as follows:

[0074]

[0075] in, Let t be the power consumption of the data center at time t. The Power Usage Effectiveness (PUE) evaluation index coefficient for data centers. Let K be the power consumption of the k-th type of server. These represent the actual charging and discharging power of the energy storage device at time t and h, respectively.

[0076] Based on the specific description of steps S12 to S14, it can be concluded that the present invention obtains the power consumption calculation rules corresponding to computing power requirements through the following method: based on the total workload of each time unit, combined with the correlation between server energy consumption and the number of active operations and the load intensity, the total power consumption of the server is calculated, and corrected using the power utilization efficiency of the data center; the algebraic difference between the charging power and discharging power of all energy storage devices is superimposed, and the summation result of the algebraic difference is added to the corrected total power consumption of the server to form a quantitative result of power consumption that matches the current computing power requirements.

[0077] Step S2, Construct a power model: Take into account the power generation capacity of adjustable and non-adjustable units, formulate the output plan curve of the generator units, and make dynamic adjustments according to the real-time operating status of the power grid.

[0078] This invention establishes stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets based on generator set type. Combining load resource parameters and regulation characteristics, a power model is constructed to obtain the total output calculation rules for the generating unit and load regulation constraint rules. On the power side related to the data center, the generator set output plan is customized the day before and flexibly adjusted during implementation, while electricity prices are calculated. On the computing side, adjustable resources are collaboratively optimized and scheduled based on power output.

[0079] Data centers have two main types of generator sets: adjustable generator sets, such as thermal power units, and non-adjustable generator sets, such as wind power units and photovoltaic power units. Assuming the data center's power nodes include p generator sets G, we will model the adjustable and non-adjustable generator sets separately.

[0080] The adjustable generator set operates within its technically permissible output range, and the output variation between adjacent time periods does not exceed its ramp-up capability limit, maintaining a stable output according to a preset plan or adjustment command. This invention refers to the stable output characteristic rule of the adjustable generator set. Assuming the adjustable generator set's power generation situation is as follows... .

[0081] The power output of non-adjustable generator sets such as wind turbines and photovoltaic power plants is based on the probabilistic distribution characteristics of natural energy resources and the segmented power characteristics of the units themselves. The probabilistic distribution characteristics of natural energy resources include the Weibull distribution of wind speed for wind turbines and the Beta distribution of solar radiation for photovoltaic power plants. The unit output is calculated by combining the predicted energy resource values ​​under the above probabilistic distributions with the segmented power characteristics of the unit. Furthermore, the output result exhibits probabilistic fluctuations with changes in natural energy resources and cannot be actively changed through adjustment commands. This invention refers to this as the probabilistic output rule for non-adjustable generator sets.

[0082] Specifically, for wind turbines, assuming the wind speed in the area where the data center is located follows a two-parameter Weibull distribution, the probability density function of wind speed v is as follows:

[0083]

[0084] in, Let be the probability density function of wind speed. For wind speed, The shape parameter describes the skewness of the distribution. This is a scale parameter that is positively correlated with the average wind speed.

[0085] The relationship between the power characteristics of a wind turbine and wind speed can be modeled as a piecewise function, as shown in the following formula:

[0086]

[0087] in, The power of the wind turbine unit. This refers to the rated output of the fan. For wind speed, To achieve the cut-in wind speed, the wind turbine will not generate electricity below this value. This is the rated wind speed; the fan will operate at full capacity once this speed is reached. To cut off the wind speed, the fan will shut down as a protective device if the wind speed exceeds this value.

[0088] Therefore, the predicted wind power output is taken as the expected value of the power characteristics of the wind turbine, as shown in the formula:

[0089]

[0090] in, For wind power output forecasting, For the expected calculation, This refers to the power output of the wind turbine.

[0091] Meanwhile, regarding the output of the photovoltaic units, it is assumed that the solar radiation received by the data center location follows a Beta distribution. The probability density function of the standardized solar irradiance r is:

[0092]

[0093] in, To standardize solar irradiance, It is a γ function. and These are the shape parameters of the Beta distribution, which can be estimated from historical irradiance data, as shown in the formula:

[0094]

[0095] in, The average of historical irradiance data. This represents the variance of historical irradiance data.

[0096] The output of a photovoltaic power station is approximately proportional to the irradiance, as shown in the formula:

[0097]

[0098] in, To contribute to photovoltaic power plants, This refers to the rated output of a photovoltaic power station under standard conditions. To standardize solar irradiance.

[0099] The output of photovoltaic power generation is also uncertain, therefore, the output of photovoltaic power is calculated based on expectations, as shown in the formula:

[0100]

[0101] in, The expected value of photovoltaic power generation, This refers to the rated output of a photovoltaic power station under standard conditions. For the expected calculation, To standardize solar irradiance.

[0102] No. Total output of each power generation unit The specific definition is as follows:

[0103]

[0104] in, This represents the total output of the l-th generator unit. To stabilize the output of the generator set, that is, to adjust the power generation of the unit, For wind power output forecasting, Forecast of photovoltaic power generation output.

[0105] S22: Model load resources, load resources The definition is as follows:

[0106]

[0107] in, A set of parameters representing load resources; Indicates the sequence number of the resource; express Expected power consumption For load resources Minimum power, For load resources Maximum power, express Maximum adjustable power, express The quantity.

[0108] right Modeling the adjustable characteristics, and analyzing the characteristic matrix. The definition is as follows:

[0109]

[0110] in, express Adjustable capability matrix; express Its adjustable capabilities. This means that r must satisfy the condition and cannot be adjusted. This means that r can be adjusted by changing the runtime. This indicates that the load resource r can be reduced or stopped directly.

[0111] The operating status of load resources at any given time is represented by the load resource status matrix. To describe:

[0112]

[0113] in, Indicates load resources The state matrix; using and This indicates whether the load resource r is currently in an operating state that is consuming electrical energy or in a shutdown state due to power failure.

[0114] This invention integrates the stable output characteristic rules of adjustable units and the probabilistic output rules of non-adjustable units, superimposes the expected power consumption, minimum power, maximum power, and maximum adjustable power parameters of load resources, and combines the load resource adjustment types identified by the characteristic matrix to establish a power model that includes generation-side output calculation logic and load-side adjustment constraint logic. This forms the total output calculation rules of the generation unit and the load adjustment constraint rules. The total output calculation rules of the generation unit are used to determine the total output composition of a single generation unit, and the load adjustment constraint rules are used to regulate the adjustment range and adjustment method of load resources.

[0115] Step S3, Joint Optimization Solution: Construct a data center computing power and power collaborative optimization scheduling model based on an iterative two-layer model, and perform interactive and iterative optimization of the two-layer elements of power and computing power in the data center to achieve bidirectional collaborative optimization of data center computing power and power.

[0116] This invention constructs an iterative two-layer model based on power consumption calculation rules, total output calculation rules for power generation units, and load regulation constraint rules. The iterative two-layer model is structured around upper-layer power model optimization and lower-layer computing power model optimization. Based on the iterative two-layer model, a closed-loop process is completed to iterate and optimize power and computing power elements, resulting in a data center computing power and power bidirectional collaborative optimization scheduling strategy for scheduling operations by computing power nodes and power nodes.

[0117] S31: Power model optimization modeling, using the power model as the upper-level model, considering the power model constraints and objective function. The goal of using the power model as the upper-level optimization model is to optimize the output plan of generating units and the scheduling of load resources while satisfying the power system operation constraints, in order to maximize the economic operation and regulation capacity of power nodes.

[0118] S311, the power model has constraints, the first of which is the power balance constraint. At any time t, the total generating power of the power nodes must equal the sum of the total load power and network losses, as shown in the formula:

[0119]

[0120] in, Let be the output of the l-th generator unit at time t. , The discharge and charging power of energy storage devices. This represents the total power consumption of the computing nodes. This refers to the power loss in the power grid.

[0121] It is also necessary to consider the upper and lower limits of generator output; the output of each generator set must be within its technical output range.

[0122]

[0123] in, The lower and upper limits of the unit's output. Let be the output of the l-th generator unit at time t. This refers to the set of generator sets involved.

[0124] The unit's output variation within adjacent time periods must meet the ramp-up capability limit:

[0125]

[0126] in, These are the uphill and downhill limits for the unit, respectively.

[0127] The charging and discharging behavior of energy storage devices must meet the following constraints:

[0128]

[0129] in, The charging power for energy storage devices, The maximum allowable charging power for energy storage devices. This refers to the discharge power of the energy storage device. This refers to the maximum allowable discharge power of the energy storage device. This is the minimum state of charge factor for energy storage devices. The state of charge of the energy storage device. This represents the maximum state of charge (SOC) of the energy storage device. Let be the state of charge of the energy storage device at time t. The charging and discharging efficiency of energy storage devices.

[0130] Adjustable loads must meet their operating characteristics:

[0131]

[0132] in, Let r be the set of parameters for the r-th load resource. Let r be the actual power consumed by the r-th load resource at time t. It represents the set of all load resources.

[0133] Furthermore, the load adjustment amount must be within its maximum adjustment capacity:

[0134]

[0135] in, Let r be the actual power consumed by the r-th load resource at time t. Let r be the actual power consumed by the r-th load resource at time t-1. This represents the maximum regulating power of the load resource.

[0136] S312 is the objective function of the power model, which aims to reduce the total cost of the power system. The total cost of the power system includes the generation cost of generator sets, the operating losses and aging costs of energy storage equipment, the cost of adjusting load compensation, and the adjustment costs related to electricity prices, as shown in the formula:

[0137]

[0138] in, Let be the power generation cost function, representing the power generation cost of the l-th generator unit at time t. This function represents the operating cost of the energy storage system, quantifying the losses and aging costs incurred by the energy storage devices during charge and discharge regulation. Let be the load regulation compensation cost function, representing the compensation cost for adjustable loads and reflecting their regulation value. This corresponds to the electricity price.

[0139] S32: Computing power model optimization modeling, using the computing power model as the lower-level model, considering the constraints and objective function of the computing power model. The goal of using the computing power model as the lower-level optimization model is to optimize the allocation of computing resources and task scheduling in the data center while meeting the service quality of computing tasks and server operation constraints, in order to achieve optimal energy efficiency and minimum operating costs for computing nodes.

[0140] S321, Constraints of the computing power model:

[0141] All real-time tasks must be completed within the current time period, and all deferred tasks must be completed before the deadline, as detailed in the formula:

[0142]

[0143]

[0144] in, Total real-time workload For the real-time workload handled by the k-th type of server, To be scheduled to a future time The amount of deferred workload to be executed. This represents the initial total amount of deferred workload.

[0145] The number of each type of server running simultaneously cannot exceed the total number, and a certain amount of redundancy must be reserved:

[0146]

[0147] in, Let be the number of k-th type servers running at time t. Let k be the total number of servers of type k. This is the capacity redundancy factor.

[0148] The actual processing load of each server must not exceed its maximum processing capacity:

[0149]

[0150] in For the k-th type of server, the total workload that needs to be handled at time t is Let be the number of k-th type servers running at time t. For the processing speed of the k-th type of server, To maximize server utilization.

[0151] The migration of real-time tasks between different computing nodes is limited by network bandwidth:

[0152]

[0153] in, Is it to migrate to another data center? Real-time workload, This represents the maximum outbound bandwidth of node d.

[0154] The workload has transmission delay and queuing delay. To guarantee quality of service, the sum of these delays should be strictly lower than the maximum allowable delay threshold for the workload. This invention, based on the M / M / 1 queuing model analysis, defines the quality of service for the workload as follows:

[0155]

[0156] in, Queue waiting time For load processing time, This represents the maximum delay time.

[0157] Energy storage devices must meet the following requirements when participating in computing power regulation:

[0158]

[0159] in, The charging power for energy storage devices, The maximum allowable charging power for energy storage devices. This refers to the discharge power of the energy storage device. This represents the maximum allowable discharge power of the energy storage device.

[0160] S322, the objective function of the computing power model aims to minimize the total operating cost of the data center, while also taking into account energy consumption costs, task migration costs, service quality costs, and energy storage and regulation costs. It is a comprehensive optimization objective, as shown in the formula:

[0161]

[0162] in, This is a function of server energy consumption cost, which is related to the power consumed by the server and the price of electricity. The migration cost coefficient per unit task. Total migration workload, This is a cost factor per unit of time delay. Even if the delay does not exceed a threshold, a longer processing time will still affect the user experience. This cost is used to quantify this impact. This is a service quality delay penalty that is activated when the total latency exceeds a threshold. This is the adjustment and compensation cost function for energy storage devices.

[0163] S33: Joint optimization solution, based on an iterative two-layer model and a genetic algorithm with an elite retention strategy, generates a new population through selection, crossover, and mutation, and performs interactive and iterative optimization of the two-layer elements of power and computing power in the data center, realizing bidirectional collaborative optimization of computing power and power in the data center, and realizing bidirectional collaborative optimized scheduling between computing power and power nodes.

[0164] The iterative optimization process includes the following steps:

[0165] Based on the power consumption calculation rules, the computing power side basic data is input, including the type of computing task, total computing volume, priority, earliest start time and latest completion time. The computing tasks are pre-arranged to generate the initial computing power node's expected power consumption curve.

[0166] Based on the total output calculation rules of the power generation unit and the load regulation constraint rules, the power side basic data is input. The power side basic data includes the daily power generation plan curve, electricity price information, and adjustable resource regulation compensation information. The initial computing node's expected electricity consumption curve is used as the basic data. The upper-level power model optimization is performed using a mixed integer linear programming algorithm to generate resource regulation plans and regulation instructions.

[0167] Based on the computing power model, the adjustment instructions are invoked, and a genetic algorithm with an elite retention strategy is used to adjust the task scheduling scheme and energy storage charging and discharging plan, generating an updated power consumption curve for the computing power node, thus completing the optimization of the lower-level computing power model.

[0168] Calculate the norm of the expected power consumption curve of the computing node generated in two consecutive iterations. If the norm is less than a set threshold or the number of iterations reaches the preset maximum number of iterations, output the collaborative optimization scheduling strategy. If the convergence condition is not met, use the updated power consumption curve as the new basic data and return to the upper-level power model optimization step to repeat the iteration until the convergence condition is met.

[0169] In one embodiment of the present invention, the iteration counter is set to... The maximum number of iterations is The specific calculation process is as follows:

[0170] S331: Input basic data, including daily power generation plan curves. Electricity price situation In addition, there are other adjustments and compensation measures for adjustable resources that can change over time.

[0171] Input the daily computing tasks J from the computing power side, where each task... The attributes include: task type Total computational workload Priority Earliest start time and latest completion time .

[0172] S332: Based on the input task information, pre-arrange the tasks to be assigned within the computing node and generate an initial estimated power consumption curve.

[0173] Task sorting and assignment: Prioritize tasks. Sort tasks j from highest to lowest efficiency and assign them to the available server with the lowest energy cost. The execution time window for task j on server k is determined. Must meet:

[0174]

[0175] For latency-sensitive tasks, the requirements are as follows: .

[0176] Initial power consumption curve calculation: Based on the allocation results of all tasks, combined with the server energy consumption model and data center PUE, the initial expected power consumption curve of the computing node is calculated.

[0177]

[0178] in, for , , This refers to the initial charging and discharging power of the energy storage device.

[0179] S333: Real-time optimization scheduling of power nodes. In iteration τ, the power node receives the electricity consumption curve uploaded by the computing node. This is used as the base load for optimized scheduling. Constraints are defined in S311, and the optimization model is defined in S312. A mixed-integer linear programming algorithm is then used to solve the problem, resulting in optimized resource adjustment plans every 15 minutes for the next 24 hours and adjustment instructions for computing nodes. .

[0180] S334: Computing node collaborative response and optimization, where computing nodes receive adjustment commands from power nodes. The system responds by adjusting task scheduling and energy storage plans. The constraints and objective function are defined in S321 and S322. A genetic algorithm with an elitist retention strategy is then used to solve the problem. First, the actual start time of the tasks, the assigned server number, and the charging / discharging state of the energy storage devices are encoded. Next, a fitness function is defined. Finally, a new population is generated through selection, crossover, and mutation. The optimal solution is then found iteratively through genetic operations. The output of this algorithm is the task scheduling scheme for the next round. and energy storage charging and discharging plan And calculate the new power consumption curve of the computing node. .

[0181] S335: Finally, the entire method is iteratively converged, and the norm of the change in the total power consumption curve of the computing node between two iterations is calculated, as shown in the formula:

[0182]

[0183] in, , For the first and the In this iteration, the computing nodes (lower-level model) calculate and upload the projected electricity consumption curve to the power nodes (upper-level model). For the first and the In each iteration, the expected power consumption of the computing node at time t.

[0184] If the convergence condition is met or iterations exceeding a certain number of rounds The method ends when the overall difference between the electricity consumption curves generated by two consecutive iterations is less than a predetermined threshold. At that time, it was assumed that a consensus had been reached between the two models of electricity and computing power, and that a steady-state solution existed, as shown in the attached figure. Figure 3As shown. Further iterations will not bring significant improvements, so computation can be stopped. The final output of the bidirectional collaborative optimization scheduling strategy for computing power and power nodes together constitutes an efficient, economical, and reliable integrated system operation scheme for computing power nodes and power nodes to perform scheduling operations.

[0185] Based on the same technical concept as the method embodiment, according to another embodiment of the present invention, a data center computing power and power collaborative optimization scheduling system is provided, comprising:

[0186] The computing power model construction module is used to distinguish between latency-sensitive and latency-insensitive workloads based on the computing task attributes carried by the data center, establish time transfer constraints for deferred workloads and spatial migration constraints for real-time workloads, determine the total workload of the data center in each time unit based on the time transfer constraints of deferred workloads, the spatial migration constraints of real-time workloads, and the spatiotemporal distribution of workloads, and construct a computing power model based on the total workload, server energy consumption characteristics, energy storage equipment operation rules, and data center power usage efficiency. The computing power model indicates the power consumption calculation rules corresponding to the current computing power demand.

[0187] The power model construction module is used to establish stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets based on generator set type. Based on the stable output characteristic rules for adjustable generator sets, the probabilistic output rules for non-adjustable generator sets, load resource parameters and regulation characteristics, the power model is constructed. The power model indicates the total output calculation rules and load regulation constraint rules of the power generation unit.

[0188] The joint optimization solution module is used to construct an iterative two-layer model based on power consumption calculation rules, total output calculation rules of power generation units, and load regulation constraint rules. The iterative two-layer model is structured with upper-layer power model optimization and lower-layer computing power model optimization. Based on the iterative two-layer model, a closed-loop process is completed to optimize power and computing power elements, resulting in a data center computing power and power bidirectional collaborative optimization scheduling strategy for scheduling operations by computing power nodes and power nodes.

[0189] It should be understood that the data center computing power and power collaborative optimization scheduling method system in this embodiment can implement all the technical solutions in the above method embodiments. The functions of each functional module can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can be referred to the relevant descriptions in the above method embodiments, which will not be repeated here.

[0190] The present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the data center computing power and power collaborative optimization scheduling method as described above.

[0191] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the data center computing power and power collaborative optimization scheduling method as described above.

[0192] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), electronic devices, 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, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0193] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that each step in the flowchart and combinations thereof 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 device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for performing a specified function in one or more processes. The processor involved in each embodiment may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a quantum computing-based data processing logic device, an artificial intelligence processor, etc., and is not limited thereto.

[0194] 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 The function specified in one or more processes.

[0195] 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 Steps of a specified function in one or more processes.

Claims

1. A data center computing power and power collaborative optimization scheduling method, characterized in that, Includes the following steps: Based on the computing task attributes carried by the data center, latency-sensitive and latency-insensitive workloads are distinguished. Time transfer constraints for delayable workloads and spatial migration constraints for real-time workloads are established respectively. Based on the time transfer constraints for delayable workloads, the spatial migration constraints for real-time workloads, and the spatiotemporal distribution of workloads, the total workload of the data center in each time unit is determined. According to the total workload, server energy consumption characteristics, energy storage equipment operation rules, and data center power usage efficiency, a computing power model is constructed. The computing power model indicates the power consumption calculation rules corresponding to the current computing power demand. Based on the generator set type, stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets are established respectively. Based on the stable output characteristic rules for adjustable generator sets, the probabilistic output rules for non-adjustable generator sets, load resource parameters and regulation characteristics, a power model is constructed. The power model indicates the total output calculation rules and load regulation constraint rules of the generator unit. Based on the power consumption calculation rules, the total output calculation rules of the power generation unit, and the load regulation constraint rules, an iterative two-layer model is constructed. The iterative two-layer model is based on the upper-layer power model optimization and the lower-layer computing power model optimization. Based on an iterative two-layer model, a closed-loop process is completed to iteratively optimize power and computing power elements by generating power consumption curves, performing power model optimization based on power consumption curves, generating adjustment instructions, performing computing power model optimization based on adjustment instructions, and updating power consumption curves. This yields a bidirectional collaborative optimization scheduling strategy for data center computing power and power, which is then used by computing power nodes and power nodes to perform scheduling operations.

2. The method according to claim 1, characterized in that, The time transfer constraints of the deferred workload include: all workloads of the deferred workload must be completed before the preset deadline, and at any single moment, the amount of unprocessed deferred workload does not exceed the initial total workload; the workload is allocated to different times for execution according to scheduling requirements within the time interval from task receipt to the deadline.

3. The method according to claim 1, characterized in that, The spatial migration constraints of the real-time workload include: the real-time workload must maintain continuous operation when migrating between different nodes; the target node must have hardware configuration compatible with the source node and sufficient free memory capacity; and the amount of data migrated in a single migration must not exceed the maximum transmission bandwidth limit of the network link between nodes.

4. The method according to claim 1, characterized in that, The determination of the total workload of the data center in each time unit based on the time transfer constraints of deferred workloads, the spatial migration constraints of real-time workloads, and the spatiotemporal distribution of workloads includes: The time allocation of deferred workloads is determined based on the time transfer constraints of deferred workloads, and the spatial distribution of real-time workloads is determined based on the spatial migration constraints of real-time workloads. Combining the time allocation of deferred workloads and the spatial distribution of real-time workloads, the total amount of deferred workloads and real-time workloads that all nodes need to process in each time unit is integrated to obtain the total workload that the data center needs to bear in each time unit. The time unit is a preset fixed time interval, and the total workload must simultaneously match the node computing power configuration and task execution requirements in each time unit.

5. The method according to claim 1, characterized in that, The computing power model is constructed based on the total workload, server energy consumption characteristics, energy storage device operating rules, and data center power efficiency. This computing power model indicates the power consumption calculation rules corresponding to the current computing power demand, including: Based on the total workload in each time unit, and combined with the correlation between server energy consumption, the number of active servers, and the intensity of the load, the total power consumption of the server is calculated using the following formula: ； in, Let K be the total power consumption of the k-th type of server. Let be the peak power of the k-th type of server. Let k be the idle power of the server. The workload that the data center needs to handle on the k-th type of servers is determined by the number of k-th type servers in the data center and the total workload of the data center. get, The number of k-th type servers currently in operation. Let k be the processing speed of the k-th type of server; The calculated total power consumption of the server is corrected by using data center power efficiency to obtain the corrected total power consumption of the server. The algebraic difference between the charging and discharging power of all energy storage devices is superimposed. The summation of the algebraic difference is added to the corrected total power consumption of the server to obtain the power consumption quantification result that matches the current computing power demand. A computing power model is then established, which indicates the power consumption calculation rules corresponding to the current computing power demand.

6. The method according to claim 1, characterized in that, The stable output characteristic rules of the adjustable unit include: the power generation output of the adjustable unit is within its own technically permissible range, and the output variation in adjacent time periods does not exceed its own ramp-up capability limit, and the unit maintains stable output according to the preset plan or adjustment command.

7. The method according to claim 1, characterized in that, The probabilistic output rules of the non-adjustable units include: the power generation output of wind turbines and photovoltaic power station units is based on the probabilistic distribution characteristics of natural energy resources and the segmented power characteristics of the units themselves. The probabilistic distribution characteristics of natural energy resources include the Weibull distribution of wind speed two parameters corresponding to wind turbines and the Beta distribution of solar radiation corresponding to photovoltaic power stations. The unit output is calculated by combining the predicted value of energy resources under the above probabilistic distribution with the segmented power characteristics of the units, and the output result shows probabilistic fluctuation characteristics with the changes of natural energy resources, which cannot be actively changed by adjustment commands.

8. The method according to claim 1, characterized in that, The power model is constructed based on the stable output characteristic rules of adjustable units, the probabilistic output rules of non-adjustable units, load resource parameters, and regulation characteristics. The power model indicates the calculation rules for the total output of the generating units and the load regulation constraint rules, including: By integrating the stable output characteristic rules of adjustable units and the probabilistic output rules of non-adjustable units, and superimposing the expected power consumption, minimum power, maximum power, and maximum adjustable power parameters of load resources, and combining the load resource adjustment types identified by the characteristic matrix, a power model is established that includes generation-side output calculation logic and load-side adjustment constraint logic. This forms the total output calculation rules for generation units and the load adjustment constraint rules. The total output calculation rules for generation units are used to determine the total output composition of a single generation unit, while the load adjustment constraint rules are used to regulate the adjustment range and adjustment method of load resources.

9. The method according to claim 1, characterized in that, The iterative two-layer model, through a closed-loop process of generating electricity consumption curves, performing power model optimization based on the electricity consumption curves, generating adjustment commands, performing computing power model optimization based on the adjustment commands, and updating the electricity consumption curves, completes the iterative optimization of power and computing power elements, including: Based on the power consumption calculation rules, the computing power side basic data is input, including the type of computing task, total computing volume, priority, earliest start time and latest completion time. The computing tasks are pre-arranged to generate the initial computing power node's expected power consumption curve. Based on the total output calculation rules of the power generation unit and the load regulation constraint rules, the power side basic data is input. The power side basic data includes the daily power generation plan curve, electricity price information, and adjustable resource regulation compensation information. The initial computing node's expected electricity consumption curve is used as the basic data. The upper-level power model optimization is performed using a mixed integer linear programming algorithm to generate resource regulation plans and regulation instructions. Based on the computing power model, the adjustment instructions are invoked, and a genetic algorithm with an elite retention strategy is used to adjust the task scheduling scheme and energy storage charging and discharging plan, generating an updated power consumption curve for the computing power node, thus completing the optimization of the lower-level computing power model. Calculate the norm of the expected power consumption curve of the computing node generated in two consecutive iterations. If the norm is less than a set threshold or the number of iterations reaches the preset maximum number of iterations, output the collaborative optimization scheduling strategy. If the convergence condition is not met, use the updated power consumption curve as the new basic data and return to the upper-level power model optimization step to repeat the iteration until the convergence condition is met.

10. The method according to claim 9, characterized in that, The upper-level power model optimization takes minimizing the total cost of the power system as the objective function. The total cost of the power system includes the power generation cost of generator sets, the operating loss and aging cost of energy storage equipment, the adjustment load compensation cost, and the adjustment cost related to electricity prices. The lower-level computing power model optimization takes minimizing the total operating cost of the data center as the objective function. The total operating cost includes server energy consumption cost, task migration cost, service quality latency penalty cost, and energy storage equipment adjustment compensation cost.

11. A data center computing power and power collaborative optimization scheduling system, characterized in that, include: The computing power model construction module is used to distinguish between latency-sensitive and latency-insensitive workloads based on the computing task attributes carried by the data center, establish time transfer constraints for deferred workloads and spatial migration constraints for real-time workloads, determine the total workload of the data center in each time unit based on the time transfer constraints of deferred workloads, the spatial migration constraints of real-time workloads, and the spatiotemporal distribution of workloads, and construct a computing power model based on the total workload, server energy consumption characteristics, energy storage equipment operation rules, and data center power usage efficiency. The computing power model indicates the power consumption calculation rules corresponding to the current computing power demand. The power model construction module is used to establish stable output characteristic rules for adjustable generator sets and probabilistic output rules for non-adjustable generator sets based on generator set type. Based on the stable output characteristic rules for adjustable generator sets, the probabilistic output rules for non-adjustable generator sets, load resource parameters and regulation characteristics, the power model is constructed. The power model indicates the total output calculation rules and load regulation constraint rules of the power generation unit. The joint optimization solution module is used to construct an iterative two-layer model based on power consumption calculation rules, total output calculation rules of power generation units, and load regulation constraint rules. The iterative two-layer model is structured with upper-layer power model optimization and lower-layer computing power model optimization. Based on an iterative two-layer model, a closed-loop process is completed to iteratively optimize power and computing power elements by generating power consumption curves, performing power model optimization based on power consumption curves, generating adjustment instructions, performing computing power model optimization based on adjustment instructions, and updating power consumption curves. This yields a bidirectional collaborative optimization scheduling strategy for data center computing power and power, which is then used by computing power nodes and power nodes to perform scheduling operations.

12. An electronic device, characterized in that, The device includes: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs, when executed by the processors, implement the steps of the data center computing power and power collaborative optimization scheduling method as described in any one of claims 1-10.

13. A 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 steps of the data center computing power and power collaborative optimization scheduling method as described in any one of claims 1-10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the data center computing power and power collaborative optimization scheduling method as described in any one of claims 1-10.

Images