Data center multi-target scheduling system considering uncertainty of new energy

Abstract

The invention belongs to the technical field of power dispatching, and discloses a data center multi-target dispatching system considering new energy uncertainty, which comprises a data center model unit, a new energy supply model unit and a multi-target optimization unit, the data center model unit is used for establishing a characterization model for energy consumption equipment, computing power load, an energy storage system and power grid power supply of a data center; the new energy supply model unit is used for establishing a representation model for new energy output power generation, and obtaining a new energy typical output scene by using an improved k-means method; and the multi-objective optimization unit is used for establishing an electric power-computing power collaborative multi-objective optimization model according to the representation model established by the data center model unit and the new energy typical output scene, and solving by using an NSGA-II algorithm to obtain an optimal scheduling strategy so as to realize optimized operation of the data center.

Description

Technical Field

[0001] This invention belongs to the field of power dispatching technology, and in particular relates to a data center multi-objective dispatching system that takes into account the uncertainty of new energy sources. Background Technology

[0002] Faced with the pressure of energy transition brought about by "dual carbon" targets, the green and low-carbon development of data centers, as high-energy-consuming systems, has become a key focus of policy guidance. Against this backdrop, optimizing data center operation strategies to reduce energy consumption and increase the absorption rate of renewable energy has become a significant challenge for data centers today.

[0003] Current research has explored ways to improve the penetration and absorption rates of new energy sources by leveraging the complementary nature of wind and solar power, thereby reducing reliance on traditional energy sources. For example, adjusting the wind and solar power ratio in terms of capacity configuration helps reduce peak-valley differences. The hybrid Copula method more accurately characterizes the complementary relationship between wind and solar power output, improving renewable energy absorption. However, considering the uncertainties of wind and solar energy, it is difficult to maintain a high level of renewable energy absorption. Alternatively, to mitigate the impact of renewable energy uncertainties on the system, a combination of stochastic programming and robust optimization is used to characterize wind and solar uncertainties, but this is prone to significant errors. Some literature generates stochastic wind and solar scenarios through the inverse transformation of cumulative distributions, and then randomly samples the output scenario set to reflect the fluctuations in output. While current research effectively handles the uncertainties of wind and solar power, it complicates the overall model, increases solution time, and introduces larger errors. To simplify the wind and solar uncertainty model, some scholars use stochastic optimization methods to generate a large number of stochastic scenarios with probability distributions. Initial stochastic scenarios can be generated using Monte Carlo sampling or Latin hypercube sampling. However, the initial scenarios contain many similar scenarios, significantly increasing the model solution time. Therefore, adopting a reasonable scenario reduction method can effectively handle the uncertainty of new energy sources and significantly improve the model solution efficiency.

[0004] Regarding data center operation strategies, scholars have conducted research on cross-regional collaboration and the temporal and spatial scheduling of computing load. For example, some studies investigate the temporal flexibility of computing load, quantifying the data center's power demand response capability as a transferable computing load. Others utilize the geographical resource differences of data centers, exploring the spatiotemporal complementarity of user task requests through scheduling to investigate the renewable energy output of data centers in different regions. Still others propose a two-layer optimization model that considers load characteristics and user-side demand response by refining the types of computing load. Finally, some studies leverage regional electricity price differences to redistribute computing load among multiple data centers to reduce overall costs. These studies have proposed various scheduling strategies to improve economic efficiency or increase the utilization rate of renewable energy. Summary of the Invention

[0005] To address the shortcomings of existing technologies and further explore the flexibility potential of data centers while optimizing data center operations and ensuring user satisfaction, this invention proposes a multi-objective scheduling system for data centers that takes into account the uncertainty of new energy sources. This invention provides the following technical solution: A multi-objective scheduling system for data centers that takes into account the uncertainty of new energy sources includes a data center model unit, a new energy supply model unit, and a multi-objective optimization unit; The data center model unit is used to establish a representation model for the energy-consuming equipment, computing load, energy storage system and power grid supply of the data center; The new energy supply model unit is used to establish a characterization model for new energy power generation, and the improved k-means method is used to obtain typical power generation scenarios of new energy. The multi-objective optimization unit is used to establish a power-computing power collaborative multi-objective optimization model based on the representation model established by the data center model unit and typical power output scenarios of new energy, and to use the NSGA-II algorithm to solve for the optimal scheduling strategy, thereby realizing the optimized operation of the data center.

[0006] Preferably, the representation model established by the data center model unit includes a data center energy-consuming equipment model, a computing load model, an energy storage system model, and a power grid supply model.

[0007] The data center energy-consuming equipment model is as follows: ; In the formula: The number of running servers. This represents the total power consumption of the data center. The power consumption of IT equipment during period t. Power consumption of non-IT devices during period t; The computing power load model is as follows: ; In the formula: The total task requirements for the data center during time period t; , These are respectively the latency-sensitive task requirements and latency-tolerant task requirements for the data center during time period t; The energy storage system model is as follows: ; In the formula: To store electricity during the operation of energy storage devices, , These are the upper and lower limits of the energy storage device's power capacity; The power grid supply model is as follows: ; In the formula: The power consumption of the data center during time period t; This is the upper limit of the power output of the power grid bus.

[0008] Preferably, the characterization model established by the new energy supply model unit includes a wind power output model and a photovoltaic power output model; The wind power output model is as follows: ; ; The probability density function of wind power output is: ; In the formula: For the actual output of the wind turbine; This refers to the actual wind speed; To cut in wind speed; To cut off the wind speed; Rated wind speed; This refers to the rated power of the fan. It is a constant; , The coefficient for the linear segment of the wind turbine power curve; This is the proportionality coefficient.

[0009] The photovoltaic output power model is as follows: ; The probability density of photovoltaic power output is: ; In the formula: To actually contribute to photovoltaic equipment; , These are the actual light radiation intensity and the rated light radiation intensity, respectively. , , , These are the actual temperature, reference temperature, ambient temperature, and nominal operating temperature of the photovoltaic panel, respectively. Rated power of photovoltaic equipment; This represents the power temperature coefficient of a photovoltaic cell. and Indicates shape parameters, Let t be the maximum photovoltaic power generation at time t; This is a gamma function.

[0010] Preferably, in the new energy supply model unit, the method for generating typical new energy output scenarios includes: Pre-determine characteristic indicators for new energy power output scenarios and preprocess the characteristic indicator data using a normalization method; The weights of each feature index are calculated using the entropy weight method; The optimal number of clusters is determined by a comprehensive evaluation using the profile coefficient method, the sum of squared errors method, and the density separation index, thereby obtaining typical power output scenarios for new energy.

[0011] Preferably, the characteristic indicators include daily output ratio, daily peak-to-valley ratio, peak output time, valley output time, and average daily output: ; In the formula: These are the daily power generation ratio, daily peak-to-valley ratio, peak power generation time, and valley power generation time for new energy sources. , These represent the peak output time and the valley output time, respectively. Contributing daily to new energy sources , These represent the maximum and minimum output values ​​of new energy sources, respectively. To make practical contributions to new energy.

[0012] Preferably, the objective function of the power-computing power collaborative multi-objective optimization model established by the multi-objective optimization unit is: ; In the formula: The objective function is one, i.e., the economic function; A collection of scenes; For specific scene serial numbers; For scene probabilities; Configure capacity-related sub-objective functions for photovoltaic, wind power, and energy storage; The annual interest rate; The service life of the equipment; , , These are functions related to photovoltaic, wind power, and energy storage capacities, respectively. Maintain relevant sub-objective functions for the data center; , , These are the annual maintenance factors for photovoltaic, wind power, and energy storage, respectively. Run relevant sub-objective functions for a single data center scenario; , , , These are the operational indicators for the energy exchange processes between photovoltaic, wind power, energy storage charging and discharging, and data centers and the power grid. , , , These are the unit interaction energy coefficients for photovoltaic, wind power, energy storage charging and discharging, and the power grid, respectively. , , These are respectively the power purchase and output from photovoltaic, wind turbine, and power grid. , These are energy storage charging and discharging output, respectively. , These refer to the energy storage charging and discharging efficiency, respectively.

[0013] Preferably, in order to absorb as much renewable energy as possible, an objective function is established. as follows: ; In the formula: , The output of photovoltaic and wind turbines respectively during time period t; , These represent the wind power and photovoltaic power abandoned by the data center during time period t.

[0014] The beneficial effects of this invention are as follows: This invention provides a multi-objective scheduling system for data centers that takes into account the uncertainty of new energy sources, and its beneficial effects are as follows: 1) Significantly enhance the capacity for renewable energy absorption; 2) Optimize data center energy consumption and computing power scheduling; 3) Improve scheduling optimization efficiency and stability; 4) Enhance the adaptability and universality of the model; 5) Effectively reduces wind and solar power curtailment rates; 6) Achieve synergy between economic efficiency and low-carbon goals. Attached Figure Description

[0015] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic block diagram of the data center architecture in an embodiment of the present invention; Figure 3 This is a schematic diagram of delay-tolerant task latency scheduling in an embodiment of the present invention; Figure 4 This is a schematic diagram of the Pareto front obtained from the multi-objective optimization scheduling solution proposed in this invention; Figure 5 This is a schematic diagram of load scheduling in the technical effect analysis of the present invention; Figure 6 This is a schematic diagram illustrating the data center computing load scheduling in the technical effect analysis of this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 This invention proposes a multi-objective scheduling system for data centers that considers the uncertainties of new energy sources, comprising a data center model unit, a new energy supply model unit, and a multi-objective optimization unit. The data center model unit establishes characterization models for the data center's energy-consuming equipment, computing load, energy storage system, and grid power supply. The new energy supply model unit establishes characterization models for new energy power generation, using an improved k-means method to obtain typical new energy output scenarios. The multi-objective optimization unit establishes a power-computing power collaborative multi-objective optimization model based on the characterization models established by the data center model unit and the typical new energy output scenarios, and uses the NSGA-II algorithm to solve for the optimal scheduling strategy, thereby achieving optimized operation of the data center. The overall system flow is illustrated below. Figure 1 As shown.

[0018] The specific functional implementation of each part of this system is described below: In this embodiment, the data center architecture is as follows: Figure 2 As shown, the data center mainly consists of distributed renewable energy power generation (wind and solar generators), energy storage systems, servers, cooling equipment, and lighting equipment. Accordingly, the representational models established for the data center model unit in this embodiment include a data center energy-consuming equipment model, a computing load model, an energy storage system model, and a power grid supply model.

[0019] 1. Data Center Energy Consumption Equipment Model Data center energy-consuming equipment is mainly divided into three categories: Information Technology (IT) equipment, cooling equipment, and lighting equipment. IT equipment, as the main carrier of computing power, undertakes the core functions of data computing and processing, accounting for 60% of energy consumption; cooling systems ensure stable equipment operation through environmental temperature control, accounting for approximately 30% of energy consumption; and basic lighting facilities, as auxiliary units, maintain an energy consumption ratio of around 10%. Since its introduction, the Power Usage Effectiveness (PUE) indicator has been recognized and applied in the data center industry both domestically and internationally. This indicator measures the ratio between the total energy consumption of a data center and the energy consumption of IT equipment, effectively characterizing the non-linear characteristics of non-IT equipment energy consumption.

[0020] Considering that the energy consumption of IT equipment mainly comes from the CPU, under a fixed CPU operating voltage and frequency, the power consumption of a server can be approximated as a linear function of the utilization rate. The specific model is shown in equation (1).

[0021] (1); In the formula: The power consumption of IT equipment during time period t; The CPU utilization rate during time period t; and This refers to the server's static power and full-load power.

[0022] Server utilization can be calculated by the total number of tasks arriving at the data center at any given time and the server's service rate. To ensure the data center's response speed, a certain margin will be left in the utilization rate.

[0023] (2); (3); In the formula: This represents the data load processed by each server in the data center during time period t. The server's service efficiency during time period t; To maximize server utilization; Calculate the performance coefficients for the server; The CPU's operating frequency during time period t.

[0024] Based on dynamic voltage and frequency regulation technology, IT power consumption can be approximated as: (4); In the formula: , These are constant coefficients.

[0025] Power consumption of non-IT equipment is estimated using the Power Usage Effectiveness (PUE) index, which is related to ambient temperature and cooling equipment efficiency. Assuming the PUE index remains constant during operation, the power consumption of non-IT equipment... for: (5); In the formula: This refers to the Power Usage Effectiveness (PUE) index for data centers.

[0026] Data center energy consumption consists of both IT equipment and non-IT equipment. Total power consumption of the data center during time period t: ;(6); In the formula: The number of running servers. The power consumption of IT equipment during period t. Power consumption of non-IT devices during period t.

[0027] 2. Computing power load model Based on users' network demands and according to task priority and response speed, data centers can categorize computing loads into latency-sensitive loads and latency-tolerant loads. (7); In the formula: The total task requirements for the data center during time period t; , These are respectively the latency-sensitive task requirements and latency-tolerant task requirements for the data center during time period t.

[0028] For latency-tolerant tasks, data centers can make appropriate adjustments to shift computing load demands over time, such as... Figure 3 As shown: (8); In the formula: For the overall task requirements of the data center; , These represent the number of tolerable tasks added and the number of tolerable tasks reduced on the server during time period t. For time period t, the server has pending delay-tolerant tasks. , These are binary variables representing the addition and reduction of tolerant tasks during time period t.

[0029] While optimizing power load, data centers must ensure Quality of Service (QoS) for users. The transmission and queuing latency of server computing load both follow the M / M / 1 queuing model and must not exceed the latency limits of the Service Level Agreement (SLA). : (9); In the formula: For the overall task requirements of the data center; Server service efficiency; This represents the number of servers running.

[0030] 3. Energy Storage System Model: The upper and lower limits of the amount of electricity that can be stored during the operation of energy storage devices are as follows: (10); In the formula: To store electricity during the operation of energy storage devices, , These represent the upper and lower limits of the energy storage device's power capacity.

[0031] The energy stored in the energy storage device is continuous in each time period. In order to ensure that the initial state of the energy storage is consistent in each operating day, the constraint that the initial and final states of the energy storage must be added.

[0032] (11); In the formula: , These refer to the charging and discharging efficiencies of energy storage devices, respectively. , These represent the charging and discharging power of the energy storage device during time period t; Rated charging and discharging power; , These are the binary variables for charging and discharging the energy storage device during time period t.

[0033] 4. Power Grid Supply Model: During normal operation, the data center needs to purchase electricity from the grid during periods of power shortage, but the power purchased is less than the grid's maximum power capacity. To avoid fluctuations in the power transmitted by the data center affecting the stable operation of the grid, selling electricity to the grid is not currently considered. (12); In the formula: The power consumption of the data center during time period t; This is the upper limit of the power output of the power grid bus.

[0034] In this embodiment, the main sources of new energy power supply are wind power and photovoltaic power generation. Accordingly, the characterization models established by the new energy supply model unit include a wind power output power model and a photovoltaic power output power model.

[0035] 1-1. Wind Power Output Model The output power of a fan is affected by wind speed, and the relationship is as follows: (13); (14); In the formula: For the actual output of the wind turbine; This refers to the actual wind speed; To cut in wind speed; To cut off the wind speed; Rated wind speed; This refers to the rated power of the fan. It is a constant; , This represents the coefficient of the linear segment of the wind turbine power curve.

[0036] The wind speed model follows a Weibull distribution, and the probability density function of wind power output can be derived from the probability density function of wind speed, specifically: Let a variable The probability density distribution is ,function Differentiable, then The probability density is: (15); In the formula: The value of the random variable representing the output power of the wind turbine; , ; for The inverse function of .

[0037] when When it is a linear function, The probability density function is: (16); In the formula: , These are the proportional coefficient and the offset constant, respectively.

[0038] Based on the above formula, the probability density function of wind power output can be derived as follows: (17); In the formula, It is a constant.

[0039] 1-2. Photovoltaic Output Power Model The output power of photovoltaic units is affected by radiation intensity The influences, and their relationships are as follows: (18); In the formula: To contribute to the actual development of photovoltaics; , These are the actual light radiation intensity and the rated light radiation intensity, respectively. , , , These are the actual temperature, reference temperature, ambient temperature, and nominal operating temperature of the photovoltaic panel, respectively. Rated power of photovoltaic equipment; This represents the power temperature coefficient of a photovoltaic cell.

[0040] The light intensity model follows a beta distribution, and the probability density function of photovoltaic output can be derived from the probability density function of light intensity, specifically: Let a variable be The probability density distribution is ,function Differentiable, then The probability density function is: (19); In the formula: The value of the photovoltaic power output random variable; , ; for The inverse function of .

[0041] when When it is a linear function, The probability density function is: (20); In the formula: , These are the proportionality coefficient and the offset constant, respectively, which are composed of parameters such as rated illumination and rated power.

[0042] Light intensity Follows a beta distribution: (twenty one); In the formula: Normalized light intensity; Contribute to photovoltaic equipment; , These are the proportional coefficient and the offset constant, respectively. This is a beta function.

[0043] Substituting the above equation into the linear probability density transformation formula, we can derive the probability density function of photovoltaic power output as follows: (twenty two); In the formula, and Indicates shape parameters, Let t be the maximum photovoltaic power generation at time t; This is a gamma function.

[0044] The above models are subject to the following power balance constraints: (twenty three) 2. In the new energy supply model unit, the methods for generating typical new energy output scenarios mainly include the following: 2-1. Pre-determine the characteristic indicators of new energy power output scenarios, and use the normalization method to preprocess the characteristic indicator data.

[0045] Feature extraction and cluster analysis of wind and solar power output scenarios are important methods for studying the uncertainty of new energy power output. This invention uses an improved k-means method to reduce the scenarios and constructs characteristic indicators for new energy power output scenarios, such as daily power output ratio, daily peak-to-valley ratio, peak power output time, valley power output time, and average daily power output. Compared with traditional average power output and rate of change indicators, the indicators selected in this invention can effectively characterize the uncertainty of wind and solar power output from the perspectives of time distribution and amplitude differences, efficiently reducing data dimensionality. The method for establishing the characteristic indicators is as follows: (twenty four); In the formula: These are the daily power generation ratio, daily peak-to-valley ratio, peak power generation time, and valley power generation time for new energy sources. , These represent the peak output time and the valley output time, respectively. Contributing daily to new energy sources , These represent the maximum and minimum output values ​​of new energy sources, respectively. To make practical contributions to new energy.

[0046] To eliminate the influence of dimensions, a normalization method is used to preprocess the data to ensure that all indicators are compared on the same dimension.

[0047] (25); In the formula: Standard values ​​for new energy output indicators; Indicator values ​​for contributing to new energy sources; and This represents the maximum or minimum value of the indicator.

[0048] 2-2. Calculate the weights of each feature index using the entropy weight method.

[0049] The specific steps for calculating the weights of the above indicators using the entropy weight method are as follows: 1) Calculate the contribution of the indicators: (26); 2) Calculate information entropy: (27); In the formula: The number of evaluation indicators.

[0050] 3) Determine the indicator weights: (28); In the formula: , The first The or the first The information entropy value corresponding to each indicator.

[0051] Based on the feature index weights, construct the clustering analysis objective function: (29); In the formula: For the first The first scenario One indicator; For the first Cluster center in similar scenarios One indicator.

[0052] 3. An improved k-means method is used for comprehensive evaluation to determine the optimal number of clusters, thereby obtaining typical power output scenarios for new energy.

[0053] To determine the optimal number of k-means clusters, the silhouette coefficient (SC), sum of squared errors (SSE), and Dunn index (DI) are used for comprehensive evaluation. The specific calculation method is as follows: 1) Silhouette coefficient: Measures the density of a sample within its cluster and its separation from the nearest cluster. The higher the coefficient, the more reasonable the sample partitioning. The mathematical model is as follows: (30); In the formula: For the sample The average distance to other samples in its cluster (cluster density); For the sample The cluster it belongs to; For the sample and The Euclidean distance between them; For the sample Average distance to the nearest cluster (inter-cluster separation); For the sample Clusters other than those mentioned above.

[0054] 2) Sum of Squared Errors: This measures the density of clusters. It calculates the sum of the squared errors between samples within a cluster and the cluster center. The smaller this index, the denser the clusters. The mathematical model is as follows: (31); In the formula: This is the sum of squared errors; The number of clusters; For the first A cluster; For clusters The mean.

[0055] 3) Density Separation Index: This measures the ratio of inter-cluster separation to intra-cluster compactness. A higher index indicates greater inter-cluster separation, higher intra-cluster compactness, and better clustering performance. The mathematical model is as follows: (32); In the formula: Density separation index; For clusters and The minimum inter-sample distance between them For clusters The maximum inter-sample distance within; For the sample With sample The distance between them.

[0056] In this embodiment, taking into account factors such as the configuration capacity and operation and maintenance of each device in the data center, the objective function of the power-computing power collaborative multi-objective optimization model established by the multi-objective optimization unit is: (33); In the formula: The objective function is one, i.e., the economic function; A collection of scenes; For specific scene serial numbers; For scenario probabilities, this invention uses one year's worth of data for simulation, with each scenario having an equal probability; Configure capacity-related sub-objective functions for photovoltaic, wind power, and energy storage; The annual interest rate; The service life of the equipment; , , These are functions related to photovoltaic, wind power, and energy storage capacities, respectively. Maintain relevant sub-objective functions for the data center; , , These are the annual maintenance factors for photovoltaic, wind power, and energy storage, respectively. Run relevant sub-objective functions for a single data center scenario; , , , These are the operational indicators for the energy exchange processes between photovoltaic, wind power, energy storage charging and discharging, and data centers and the power grid. , , , These are the unit interaction energy coefficients for photovoltaic, wind power, energy storage charging and discharging, and the power grid, respectively. , , These are respectively the power purchase and output from photovoltaic, wind turbine, and power grid. , These are energy storage charging and discharging output, respectively. , These refer to the energy storage charging and discharging efficiency, respectively.

[0057] At the same time, in order to absorb as much renewable energy as possible, an objective function is established. as follows: (34); In the formula: , The output of photovoltaic and wind turbines respectively during time period t; , These represent the wind power and photovoltaic power abandoned by the data center during time period t.

[0058] The solution method is as follows: In multi-constraint, multi-objective optimization scenarios, both NSGA-II and MOEA / D are widely used solution methods. Existing research, using the ZDT test function, has verified that the Pareto front obtained by NSGA-II has a uniform shape and closely matches the true front, while the Pareto front of MOEA / D exhibits numerous discontinuities. Furthermore, as the number of decision variables increases, the NSGA-II algorithm requires significantly fewer iterations to find the optimal solution compared to the MOEA / D algorithm.

[22] Therefore, this invention uses the NSGA-II algorithm for solving the problem.

[0059] The NSGA-II algorithm, as a highly efficient multi-objective optimization method, is based on a non-dominated sorting mechanism and an elite retention strategy. It achieves multi-objective collaborative optimization by obtaining a Pareto front solution set. This algorithm rapidly eliminates inferior solutions through non-dominated hierarchical elimination and ensures the distribution and diversity of the solution set by combining a crowding index, effectively avoiding the tendency of traditional algorithms to get trapped in local optima in multi-objective trade-offs. Furthermore, it does not require converting multiple objectives into single-objective weighted processing, and can intuitively present the competitive relationships between objectives, demonstrating good performance in solving scheduling problems.

[0060] NSGA-II solution steps: The main steps of the algorithm are as follows: Step 1: Generate the initial population .

[0061] Step 2: Fast Non-Dominated Sort. Before performing the selection operation, the population is first stratified according to the level of non-dominated individual solutions to obtain the parent individuals. .

[0062] Step 3: Genetic operators (selection, crossover, and mutation) generate new individuals. .

[0063] Step 4, Elite Strategy: First, prioritize the parent generation... and genetic operators merge into one population population The number of individuals becomes 2 Secondly, regarding the population Perform non-dominated sorting and calculate the corresponding crowding distance; finally, select individuals one by one according to their level until the total number of individuals reaches [the target value]. This makes it the parent population for a new round of evolution. .

[0064] Step 5: Repeat steps 3-4 until the iteration termination condition is met.

[0065] After completing the cluster analysis, a scenario optimization index based on the Spearman correlation coefficient is constructed to evaluate the consistency and correlation between the two sets of power output scenarios in terms of their changing trends. The formula is as follows: (35); In the formula: Ranking difference sets for two new energy power output scenarios; The number of elements contributing to new energy.

[0066] New energy power output scenarios It can be represented in matrix form: (36); In the formula: A collection of scenarios contributing to new energy; For the scene exist The output of new energy sources during certain periods.

[0067] The Spearman correlation coefficient was used to calculate the sum of correlation coefficients between each output scenario in the scenario set and other scenarios, and the scenario with the highest sum of correlation coefficients was selected as the typical scenario. Typical scenario selection criteria. The definition is as follows: (37); In the formula: For new energy scenarios With Scene The correlation coefficients are shown in Table 1 for wind power cluster analysis and Table 2 for photovoltaic cluster analysis results.

[0068] Table 1 Table 2 Example 2 This embodiment provides an effect analysis of the technical solution of the present invention. The data center latency-tolerant tasks account for 40% of the total, and 25,000 large model tasks are migrated in at 0:00 to simulate a data center training task.

[0069] The Pareto front obtained by the multi-objective optimization scheduling solution proposed in this invention is as follows: Figure 4 As shown in the figure, all solutions are optimal. The figure shows a negative correlation between IDC operating costs and the system's wind and solar curtailment rate; that is, as the wind and solar curtailment rate decreases, operating costs gradually increase. Based on the different decisions regarding the two objectives, the weights of objective 1 and objective 2 are selected as 0.4 and 0.6 respectively. The congestion index in the NSGA-II algorithm is calculated, and... Figure 4 The central pentagonal point is the optimal solution of this invention.

[0070] From the perspectives of power and computing power, the data center has fewer task scheduling requests, a gradual reduction trend, and a user satisfaction score of 0.983. Its load scheduling is as follows: Figure 5 As shown.

[0071] Data center power consumption, such as Figure 5 As shown, the difference between wind and solar power output and data center supply and demand is relatively small. This allows for the maximum absorption of new energy sources and a reduction in grid power output by leveraging the complementary characteristics of wind and solar power and the flexibility of computing load. To extend the service life of energy storage devices, the charge and discharge thresholds should be increased to avoid frequent charging and discharging, which could reduce battery cycle life.

[0072] Data center computing load scheduling status as follows Figure 6 As shown, during periods of high electricity prices and scarce renewable energy sources, tasks are migrated to periods of low electricity prices and abundant renewable energy sources, achieving a dual response in terms of computing power load and power consumption.

[0073] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A multi-objective scheduling system for data centers that takes into account the uncertainty of new energy sources, characterized in that, It includes data center model units, new energy supply model units, and multi-objective optimization units; The data center model unit is used to establish a representation model for the energy-consuming equipment, computing load, energy storage system and power grid supply of the data center; The new energy supply model unit is used to establish a characterization model for new energy power generation, and to obtain typical new energy power generation scenarios using the improved k-means method. The multi-objective optimization unit is used to establish a power-computing power collaborative multi-objective optimization model based on the representation model established by the data center model unit and typical power output scenarios of new energy, and to use the NSGA-II algorithm to solve for the optimal scheduling strategy, thereby realizing the optimized operation of the data center.

2. The system according to claim 1, characterized in that, The representation models established by the data center model unit include data center energy-consuming equipment models, computing load models, energy storage system models, and power grid power supply models. The data center energy-consuming equipment model is as follows: ; In the formula: The number of running servers. The total power consumption of the data center during time period t. The power consumption of IT equipment during period t. Power consumption of non-IT devices during period t; The computing power load model is as follows: ; In the formula: The total task requirements for the data center during time period t; , These are respectively the latency-sensitive task requirements and latency-tolerant task requirements for the data center during time period t; The energy storage system model is as follows: ; In the formula: To store electricity during the operation of energy storage devices, , These are the upper and lower limits of the energy storage device's power capacity; The power grid supply model is as follows: ; In the formula: The power consumption of the data center during time period t; This is the upper limit of the power output of the power grid bus.

3. The system according to claim 1, characterized in that, The characterization models established by the new energy supply model unit include wind power output power model and photovoltaic power output power model; The wind power output model is as follows: ； ； The probability density function of wind power output is: ； In the formula: For the actual output of the wind turbine; This refers to the actual wind speed; To cut in wind speed; To cut off the wind speed; Rated wind speed; This refers to the rated power of the fan. It is a constant; , The coefficient for the linear segment of the wind turbine power curve; This is the proportionality coefficient; The photovoltaic output power model is as follows: ; The probability density of photovoltaic power output is: ； In the formula: To actually contribute to photovoltaic equipment; , These are the actual light radiation intensity and the rated light radiation intensity, respectively. , , , These are the actual temperature, reference temperature, ambient temperature, and nominal operating temperature of the photovoltaic panel, respectively. Rated power of photovoltaic equipment; The power temperature coefficient of a photovoltaic cell. and Indicates shape parameters, Let t be the maximum photovoltaic power generation at time t; This is a gamma function.

4. The system according to claim 3, characterized in that, The method for generating typical new energy output scenarios in the new energy supply model unit includes: Pre-determine characteristic indicators for new energy power output scenarios and preprocess the characteristic indicator data using a normalization method; The weights of each feature index are calculated using the entropy weight method; The optimal number of clusters is determined by a comprehensive evaluation using the profile coefficient method, the sum of squared errors method, and the density separation index, thereby obtaining typical power output scenarios for new energy.

5. The system according to claim 4, characterized in that, The characteristic indicators include daily output ratio, daily peak-to-valley ratio, peak output time, valley output time, and average daily output: ； In the formula: These are the daily power generation ratio, daily peak-to-valley ratio, peak output time, valley output time, and average daily output of new energy sources. , These represent the peak output time and the valley output time, respectively. Contributing daily to new energy sources , These represent the maximum and minimum output values ​​of new energy sources, respectively. To make practical contributions to new energy.

6. The system according to claim 1, characterized in that, The objective function of the power-computing power collaborative multi-objective optimization model established by the multi-objective optimization unit is: ； In the formula: The objective function is one, i.e., the economic function; A collection of scenes; For specific scene serial numbers; For scene probabilities; Configure capacity-related sub-objective functions for photovoltaic, wind power, and energy storage; The annual interest rate; The service life of the equipment; , , These are functions related to photovoltaic, wind power, and energy storage capacities, respectively. Maintain relevant sub-objective functions for the data center; , , These are the annual maintenance factors for photovoltaic, wind power, and energy storage, respectively. Run relevant sub-objective functions for a single data center scenario; , , , These are the operational indicators for the energy exchange processes between photovoltaic, wind power, energy storage charging and discharging, and data centers and the power grid. , , , These are the unit interaction energy coefficients for photovoltaic, wind power, energy storage charging and discharging, and the power grid, respectively. , , These are respectively the power purchase and output from photovoltaic, wind turbine, and power grid. , These are energy storage charging and discharging output, respectively. , These refer to the energy storage charging and discharging efficiency, respectively.

7. The system according to claim 6, characterized in that, To absorb a high proportion of renewable energy, an objective function is established. as follows: ； In the formula: , The output of photovoltaic and wind turbines respectively during time period t; , These represent the wind power and photovoltaic power abandoned by the data center during time period t.

Images