Big data-based data center computing power and power collaborative regulation system

Abstract

The application relates to the technical field of electric power energy data management, and particularly discloses a data center computing power and electric power collaborative regulation system based on big data, which comprises a data sensing module, a fusion modeling module, a simulation analysis module and a collaborative regulation module. The scheme constructs a unified computing power-electric power fusion model, establishes an accurate and quantitative bridge between computing power scheduling instructions and electric power load changes, and accurately reflects the influence of environmental temperature and computing power task mixed mode on energy consumption through dynamic CLF modeling and data driving calibration, so that the accuracy of electric power load prediction is effectively improved; through high-precision prediction and simulation, the adjustable potential of the data center is quantified, the power consumption behavior is dynamically adjusted according to the power grid signal, and the intelligent optimization algorithm is used to automatically optimize among multiple targets, so that the operation of the power grid is stabilized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power energy data management, in particular to a data center computing power and electric power collaborative regulation system based on big data. BACKGROUND

[0002] With the vigorous development of cloud computing, artificial intelligence and digital economy, the computing power demand and energy consumption of global data centers have shown explosive growth. Data centers have become key energy consumption units, and the electricity cost accounts for a high proportion of their operating costs. At the same time, the intensive power consumption of data centers also brings pressure to the stable operation of regional power grids, especially during peak electricity consumption periods. Traditional computing power models and power models are heterogeneous, one based on task queues and resource utilization, and the other based on physical power consumption and thermodynamics. They cannot directly communicate with each other, resulting in the inability to assess the impact of computing power scheduling decisions on power load from a global perspective. The regulation of traditional solutions is mostly passive and responsive, and cannot proactively and actively involve computing power as a flexible resource in the DSR (demand side response) of the power grid. SUMMARY

[0003] In view of the above, in order to overcome the defects of the prior art, the present application provides a data center computing power and electric power collaborative regulation system based on big data, which is aimed at the problem that traditional computing power models and power models are heterogeneous, one based on task queues and resource utilization, and the other based on physical power consumption and thermodynamics. They cannot directly communicate with each other, resulting in the inability to assess the impact of computing power scheduling decisions on power load from a global perspective. The present solution constructs a unified computing power-power fusion model, establishes an accurate and quantitative bridge between computing power scheduling instructions and power load changes, and accurately reflects the impact of environmental temperature and computing power task mixed mode on energy consumption through dynamic CLF (cooling load factor) model and data-driven calibration, effectively improving the accuracy of power load prediction. In view of the problem that the regulation of traditional solutions is mostly passive and responsive, and cannot proactively and actively involve computing power as a flexible resource in the DSR of the power grid, the present solution quantifies the adjustable potential of the data center through high-precision prediction and simulation, dynamically adjusts the electricity consumption behavior according to the power grid signal, and automatically optimizes between multiple targets through intelligent optimization algorithm, stabilizing the operation of the power grid.

[0004] The data center computing power and electric power collaborative regulation system based on big data provided by the present application comprises a data perception module, a fusion modeling module, a simulation analysis module and a collaborative regulation module.

[0005] The data perception module collects the operating data and energy consumption data of the internal IT equipment of the data center, as well as external power grid data and environmental data.

[0006] The fusion modeling module fuses modeling through data mechanism, and constructs a data center computing power-power fusion model describing the internal correlation between computing power tasks and power load;

[0007] The simulation analysis module simulates different computing power task scheduling schemes based on the computing power-power fusion model, predicts the temporal and spatial variation of power load caused by the schemes, and generates a complete simulation evaluation report;

[0008] The cooperative regulation module uses an optimization algorithm to solve the optimal computing power-power cooperative regulation strategy based on the simulation evaluation report, and jointly controls the computing power resources and power resources.

[0009] Further, the fusion modeling module includes a parameter calculation unit, a power load modeling unit, a computing power load modeling unit, a data driving unit, and a model fusion unit;

[0010] The parameter calculation unit inputs the internal and external data collected by the data perception module, calculates PUE (power usage effectiveness), sets initial parameters, constructs a dynamic CLF model, and uses the following formula:

[0011] ;

[0012] ;

[0013] In the formula, PUE represents power usage effectiveness, Ptotal represents the total power consumption of the data center, Pit represents the total power consumption of IT equipment, C represents the cooling load coefficient, T represents the ambient temperature, R represents the total load rate of IT equipment, T_s represents the ambient temperature sensitivity coefficient, L_s represents the IT load sensitivity coefficient, C_base represents the basic cooling coefficient;

[0014] The power load modeling unit describes the total power consumption of the data center power load side from a macro perspective based on the operation data and energy consumption data of the internal IT equipment of the data center, constructs a power load mechanism model, and the model form is as follows:

[0015] ;

[0016] ;

[0017] In the formula, P_base represents the infrastructure energy consumption;

[0018] The computing power load modeling unit combines data center operation data, analyzes the operation state of the server when processing different computing power tasks, establishes the relationship between server power consumption and computing power load from a micro level, constructs a computing power load mechanism model, and the model form is as follows:

[0019]

[0020] In the formula, Ptotal represents the total power consumption of the server, Pidle represents the idle power consumption of the server without computing tasks, Pfull represents the full load power consumption of the server, CPU represents the central processing unit utilization rate;

[0021] The data driving unit constructs a data-driven model based on a gradient boosting decision tree algorithm , uses historical operation data of the data center IT equipment to construct a high-dimensional feature vector, extracts the associated features between the high-dimensional feature vector and the server power consumption through the data-driven model, and the model form is as follows:

[0022]

[0023] In the formula, Ppred represents the predicted server power consumption, D represents the data-driven model, X represents the high-dimensional feature vector;

[0024] The model fusion unit calibrates the parameters of the power load mechanism model based on the data-driven model to form a data center computing power-power fusion model .

[0025] Further, the model fusion unit comprises a load characterization subunit, a parameter calibration subunit and a model fusion subunit.

[0026] The load characterization subunit analyzes the distribution of the high-dimensional feature vector to generate a context vector representing the current computing power load characteristics of the data center .

[0027] The parameter calibration subunit constructs a calibration model based on a shallow neural network , inputs the context vector and external environment data, and uses the calibration model to output the optimal parameters of the dynamic CLF model, so that the dynamic CLF model is adapted to the current specific computing power task mixed mode of the data center, and the formula is as follows:

[0028]

[0029] In the formula, X represents the context vector, C represents the calibration model,​​​ optimal parameters of the dynamic CLF model;

[0030] The model fusion subunit substitutes the optimal parameters into the dynamic CLF model to form and output a data center computing power-power fusion model The updated power load mechanism model is expressed as follows:

[0031] .

[0032] Further, the simulation analysis module includes a simulation scenario definition unit, a simulation deduction unit, a power load calculation unit, and a result quantification unit.

[0033] The simulation scenario definition unit obtains a candidate computing power task scheduling scheme set of the data center , and an external environment data sequence in the future period of time ; the candidate scheduling scheme in the candidate computing power task scheduling scheme set is a structured instruction set, describing the redistribution of computing power tasks in space and time.

[0034] The simulation deduction unit simulates each candidate scheduling scheme, and in each simulation time step, updates the task load of the server according to the real-time running data of the internal IT equipment, and calculates the computing power index at each moment.

[0035] The power load calculation unit inputs the task load at the current moment and the external environment data sequence into the computing power-power fusion model to calculate the total power consumption of the data center at this moment, and generates a total power consumption curve for each candidate scheduling scheme, using the following formula:

[0036] ;

[0037] In the formula, represents the total power consumption curve of the simulation, represents the computing power task scheduling scheme, represents the external environment data sequence.

[0038] The result quantification unit deeply analyzes the total power consumption curve output by the simulation to generate multi-dimensional evaluation indexes, and generates a complete simulation evaluation report for each candidate scheduling scheme.

[0039] Further, the collaborative regulation module includes a target function construction unit, a constraint condition modeling unit, an optimization algorithm solving unit, and a strategy output unit.

[0040] The target function construction unit performs multi-objective optimization based on the simulation evaluation report, converts the multi-dimensional evaluation indexes into a comprehensive target function, and the optimization target is to minimize the total cost, using the following formula:

[0041] ;

[0042] In the formula, represents the minimum value operation on the comprehensive target function, represents the comprehensive target function, represents the minimization function, represents the electricity cost obtained by simulation, represents the carbon emissions obtained by simulation, represents the additional income obtained by responding to the grid signal, represents the economic cost weight, represents the environmental cost weight, represents the grid income weight.

[0043] The constraint condition modeling unit constructs the policy constraints of the data center, and the constraints include computing power resource constraints, power capacity constraints and other constraints.

[0044] The optimization algorithm solving unit iteratively updates all candidate scheduling scheme combinations using a genetic algorithm, calculates the comprehensive target function value, and converges to obtain a set of optimal computing power task scheduling schemes. ;

[0045] The policy output unit outputs the optimal computing power task scheduling scheme, which is decomposed into an executable computing power scheduling instruction set and a power control instruction set, and is respectively sent to the computing power scheduling engine and the power control engine of the data center.

[0046] The beneficial effects obtained by the above-mentioned scheme are as follows:

[0047] (1) The traditional computing power model and the power model are heterogeneous, one is based on task queue and resource utilization, and the other is based on physical power consumption and thermodynamics, and the two cannot directly communicate, resulting in the problem that the influence of computing power scheduling decision on power load cannot be evaluated from a global perspective. The scheme constructs a unified computing power-power fusion model, establishes an accurate and quantitative bridge between computing power scheduling instructions and power load changes, and accurately reflects the influence of environmental temperature and computing power task mixed mode on energy consumption through dynamic CLF model and data-driven calibration, effectively improving the accuracy of power load prediction.

[0048] (2) The traditional scheme is mostly passive and responsive, and cannot prospectively and actively use computing power as a flexible resource to participate in the DSR of the power grid. The scheme quantifies the adjustable potential of the data center through high-precision prediction and simulation, dynamically adjusts the power consumption behavior according to the grid signal, and automatically optimizes between multiple targets through an intelligent optimization algorithm, stabilizing the operation of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 A schematic diagram of the big data-based data center computing power and power collaborative regulation system according to the present application is shown in FIG. 1.

[0050] Figure 2 A schematic diagram of the fusion modeling module is shown in FIG. 2.

[0051] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application, and explain the technical solutions of the present application, but do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of the present application.

[0053] Embodiment one, refer to Figure 1 The big data-based data center computing power and power collaborative regulation system according to the present application comprises a data perception module, a fusion modeling module, a simulation analysis module, and a collaborative regulation module.

[0054] The data perception module collects the operation data and energy consumption data of the internal IT equipment of the data center, as well as external power grid data and environmental data.

[0055] The fusion modeling module constructs a data center computing power-power fusion model describing the internal correlation between computing power tasks and power loads through data mechanism fusion modeling.

[0056] The simulation analysis module simulates different computing power task scheduling schemes based on the computing power-power fusion model, predicts the temporal and spatial changes of power loads caused by the schemes, and generates a complete simulation evaluation report.

[0057] The collaborative regulation module uses an optimization algorithm to solve the optimal computing power-power collaborative regulation strategy based on the simulation evaluation report, and jointly controls the computing power resources and the power resources.

[0058] Embodiment two, refer to Figure 1 The data perception module in the present embodiment is based on the above-mentioned embodiment.

[0059] The operation data of the internal IT equipment includes CPU utilization, memory utilization, disk I / O rate, network bandwidth utilization, GPU utilization, task queue length, process / service list, virtual machine / container quantity and state.

[0060] The energy consumption data includes: server cabinet power consumption, single server power consumption, storage device power consumption, network device power consumption, total refrigeration system power consumption, power supply system loss, total data center input power consumption;

[0061] The external power grid data includes: time-of-use electricity price, real-time electricity price, demand side response incentive signal, peak demand warning, carbon emission factor;

[0062] The environmental data includes: outdoor dry bulb temperature, outdoor wet bulb temperature, relative humidity, atmospheric pressure, wind speed and direction, weather forecast data.

[0063] Embodiment three, see Figure 1 and Figure 2 , the embodiment is based on the above-mentioned embodiment, the fusion modeling module includes a parameter calculation unit, a power load modeling unit, an algorithm load modeling unit, a data driving unit and a model fusion unit;

[0064] The parameter calculation unit inputs the internal and external data collected by the data perception module, calculates the PUE, sets the initial parameters, constructs the dynamic CLF model, collects the historical operation data samples of the data center, and fits the environmental temperature sensitivity coefficient, the IT load sensitivity coefficient and the basic refrigeration coefficient through mean value calculation. The initial values are 0.01, 0.005 and 0.1 respectively, and the used formula is as follows:

[0065] ;

[0066] ;

[0067] In the formula, represents the power usage effectiveness, represents the total power consumption of the data center, represents the total power consumption of the IT equipment, represents the refrigeration load coefficient, represents the environmental temperature, represents the total load rate of the IT equipment, represents the environmental temperature sensitivity coefficient, represents the IT load sensitivity coefficient, represents the basic refrigeration coefficient;

[0068] The power load modeling unit is based on the operation data and energy consumption data of the internal IT equipment of the data center, describes the total power consumption of the power load side of the data center from a macroscopic perspective, constructs a power load mechanism model, and the model form is as follows:

[0069] ;

[0070] ;

[0071] In the formula, representing the infrastructure energy consumption;

[0072] The computing power load modeling unit combines data center operation data, analyzes the operation state of the server when processing different computing power tasks, establishes the relationship between server power consumption and computing power load from a micro level, and constructs a computing power load mechanism model. The model form is as follows:

[0073] ;

[0074] In the formula, representing the total power consumption of the server, representing the idle power consumption of the server without computing tasks, representing the full load power consumption of the server, representing the central processing unit utilization rate;

[0075] The data driving unit uses the XGBoost framework, uses the historical operation data of the data center IT equipment in the past 3 to 6 months to construct a high-dimensional feature vector, and randomly divides it into a training set and a validation set according to a ratio of 7:3. The predicted server power consumption is iteratively trained on the training set, and the model loss is calculated on the validation set. When the model loss does not decrease for 10 consecutive rounds, the training is terminated, and the model at this time is saved as the data-driven model , which predicts the server power consumption, and is used as follows:

[0076] ;

[0077] In the formula, representing the predicted server power consumption, representing the data-driven model, representing the high-dimensional feature vector;

[0078] The model fusion unit calibrates the parameters of the power load mechanism model based on the data-driven model to form a data center computing power-power fusion model .

[0079] In Example Four, refer to Figure 1 and Figure 2 , based on the above-mentioned embodiments, in the data driving unit, the high-dimensional feature vector includes:

[0080] Core resource utilization rate features: central processing unit utilization rate, memory utilization rate, disk I / O throughput, disk I / O operation frequency, network I / O throughput;

[0081] Computing task semantic features: task type encoding, process / service name;

[0082] Hardware architecture and state features: CPU core running frequency, CPU core voltage, number of instructions per clock cycle, cache hit rate, hardware performance counter events;

[0083] System level and context features: system load, number of concurrent threads, time features.

[0084] Embodiment five, refer to Figure 1 and Figure 2 , this embodiment is based on the above-mentioned embodiments, the model fusion unit comprises a load characterization subunit, a parameter calibration subunit and a model fusion subunit;

[0085] The load characterization subunit analyzes the distribution of the high-dimensional feature vector to generate a context vector representing the current computing power load characteristics of the data center ;

[0086] The parameter calibration subunit adopts a fully connected feedforward neural network to construct a calibration model , the number of neurons in the input layer is set to the same dimension as the context vector , 1 hidden layer is set, the hidden layer uses ReLU activation function, 3 neurons are set in the output layer, respectively corresponding to the three optimal parameters of the dynamic CLF model calibration , the formula used is as follows:

[0087] ;

[0088] In the formula, represents the context vector, represents the calibration model, represents the optimal parameters of the dynamic CLF model;

[0089] The model fusion subunit substitutes the optimal parameters into the dynamic CLF model to form and output a data center computing power-power fusion model , the updated power load mechanism model is represented as follows:

[0090] .

[0091] By performing the operation, for the problem that the traditional computing power model and the power model are heterogeneous, one is based on task queue and resource utilization, and the other is based on physical power consumption and thermodynamics, and the two cannot directly communicate, resulting in the inability to evaluate the influence of computing power scheduling decision on power load from a global perspective, the present scheme constructs a unified computing power-power fusion model, establishes an accurate and quantitative bridge between computing power scheduling instructions and power load changes, and through dynamic CLF model and data-driven calibration, accurately reflects the influence of environmental temperature and computing power task mixed mode on energy consumption, effectively improves the accuracy of power load prediction.

[0092] Embodiment six, refer to Figure 1 , this embodiment is based on the above embodiment, the simulation analysis module includes simulation scenario definition unit, simulation deduction unit, power load calculation unit and result quantization unit;

[0093] The simulation scenario definition unit obtains a candidate computing power task scheduling scheme set of the data center , and an external environment data sequence in the next 24 hours ; the candidate scheduling scheme in the candidate computing power task scheduling scheme set is a structured instruction set, which describes the redistribution of computing power tasks in space and time, including time migration, space migration, load consolidation and performance dynamic adjustment; the external environment data sequence includes power grid time-of-use price curve, demand side response incentive signal, future air temperature curve, and renewable energy predicted output curve;

[0094] The simulation deduction unit simulates each candidate scheduling scheme, and in each simulation time step, updates the task load of the server according to the real-time running data of the internal IT equipment, calculates the computing power index at each time, and the computing power index includes total CPU utilization and average memory usage;

[0095] The power load calculation unit inputs the task load at the current time and the external environment data sequence into the computing power-power fusion model, calculates the total power consumption of the data center at this time, generates a total power consumption curve for each candidate scheduling scheme, and the formula is as follows:

[0096] ;

[0097] In the formula, represents the total power consumption curve of the simulation, represents the computing power task scheduling scheme, represents the external environment data sequence;

[0098] The result quantization unit deeply analyzes the total power consumption curve output by the simulation, generates multi-dimensional evaluation indexes, including power load space-time transfer quantization, adjustable potential evaluation, economic and environmental pre-evaluation, and generates a complete simulation evaluation report for each candidate scheduling scheme.

[0099] Embodiment seven, refer to Figure 1 , this embodiment is based on the above embodiment, the cooperative regulation module includes a target function construction unit, a constraint condition modeling unit, an optimization algorithm solving unit and a strategy output unit;

[0100] The target function construction unit performs multi-objective optimization based on the simulation evaluation report, converts the multi-dimensional evaluation indexes into a comprehensive target function, the optimization target is to minimize the total cost, and economic cost weight, environmental cost weight and grid benefit weight are set according to the multi-dimensional evaluation indexes, the value range is between 0 and 1, and the normalization condition is met , the comprehensive target function formula is as follows:

[0101] ;

[0102] In the formula, min represents the minimum value operation on the comprehensive target function, comprehensive target function, minimization function, represents the electricity cost obtained by simulation, represents the carbon emission obtained by simulation, represents the additional income obtained by responding to the grid signal, economic cost weight, environmental cost weight, grid benefit weight;

[0103] The constraint condition modeling unit constructs the strategy constraint of the data center, and the constraint conditions include computing power resource constraint, power capacity constraint and other constraints;

[0104] The optimization algorithm solving unit adopts genetic algorithm to iteratively update all candidate scheduling scheme combinations, adopts real number coding, the population size is 100 individuals, each individual represents a complete computing power task scheduling scheme, adopts tournament selection method, the scale is 3, simulates binary crossover, the crossover probability is 0.9, and adopts polynomial mutation operation, when the optimal solution is continuously improved for 50 generations and the improvement amplitude is less than 1e-6, the iteration is stopped, and a group of optimal computing power task scheduling schemes are obtained ;

[0105] The strategy output unit outputs the optimal computing power task scheduling scheme, decomposes it into an executable computing power scheduling instruction set and a power control instruction set, and respectively sends them to the computing power scheduling engine and the power control engine of the data center.

[0106] Through the operation, the regulation and control of the traditional scheme is passive and responsive, and cannot prospectively and actively participate in the DSR of the grid as a flexible resource. The present scheme quantifies the adjustable potential of the data center through high-precision prediction and simulation, dynamically adjusts the power consumption behavior according to the grid signal, automatically optimizes between multiple targets through an intelligent optimization algorithm, and stabilizes the operation of the grid.

[0107] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other present or future technologies can provide. Specifically, it is contemplated that some embodiments of the present application can be practiced without the specific details set forth in the preceding description.

[0108] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the spirit and scope of the application, which is defined by the appended claims and their equivalents.

[0109] The above description of the application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the application, without creative design, similar structure and embodiments of the technical solution should belong to the protection scope of the application.

Claims

1. A data center computing power and power coordinated control system based on big data, characterized in that: It includes a data perception module, a fusion modeling module, a simulation analysis module, and a collaborative control module; The data sensing module collects operational and energy consumption data of IT equipment inside the data center, as well as external power grid and environmental data. The fusion modeling module constructs a data center computing power-power fusion model that describes the intrinsic relationship between computing power tasks and power load through data mechanism fusion modeling. The simulation analysis module is based on the computing power-power fusion model to simulate different computing power task scheduling schemes, predict the resulting spatiotemporal changes in power load, and generate a complete simulation evaluation report. The coordinated control module uses an optimization algorithm based on the simulation evaluation report to solve for the optimal computing power-electricity coordinated control strategy, and performs joint control of computing power resources and electricity resources. The fusion modeling module includes a parameter calculation unit, a power load modeling unit, a computing power load modeling unit, a data driving unit, and a model fusion unit; The parameter calculation unit takes in the internal and external data collected by the data sensing module, calculates the PUE, sets the initial parameters, and constructs a dynamic CLF model. The power load modeling unit describes the total power consumption on the power load side of the data center from a macro perspective, based on the operating data and energy consumption data of the IT equipment inside the data center, and constructs a power load mechanism model. The computing load modeling unit combines data center operation data to analyze the operating status of the server when processing different computing tasks, establishes the relationship between server power consumption and computing load from a micro level, and constructs a computing load mechanism model. The data-driven unit constructs a data-driven model based on the gradient boosting decision tree algorithm and uses historical operating data of data center IT equipment to construct a high-dimensional feature vector. The model fusion unit calibrates the parameters of the power load mechanism model based on the data-driven model to form a data center computing power-power fusion model; The model fusion unit includes a load characterization subunit, a parameter calibration subunit, and a model fusion subunit; The load characterization subunit analyzes the distribution of high-dimensional feature vectors and generates a context vector characterizing the current computing load characteristics of the data center. The parameter calibration subunit constructs a calibration model based on a shallow neural network, takes context vectors and external environment data as input, and uses the calibration model to output the optimal parameters of the dynamic CLF model. The model fusion subunit substitutes the optimal parameters into the dynamic CLF model to form and output the data center computing power-power fusion model and update the power load mechanism model. The simulation analysis module includes a simulation scenario definition unit, a simulation deduction unit, a power load calculation unit, and a result quantification unit. The simulation scenario definition unit obtains a set of candidate computing power task scheduling schemes for the data center, as well as a sequence of external environment data for a future period of time; The simulation and deduction unit simulates each candidate scheduling scheme. Within each simulation time step, it updates the server's task load based on the real-time operating data of the internal IT equipment and calculates the computing power index at each moment. The power load calculation unit inputs the current task load and external environment data sequence into the computing power-power fusion model to calculate the total power consumption of the data center at that moment and generate a total power consumption curve for each candidate scheduling scheme. The result quantization unit performs in-depth analysis on the total power consumption curve of the simulation output, generates multi-dimensional evaluation indicators, and generates a complete simulation evaluation report for each candidate scheduling scheme. The collaborative control module includes an objective function construction unit, a constraint modeling unit, an optimization algorithm solving unit, and a strategy output unit; The objective function construction unit performs multi-objective optimization based on the simulation evaluation report, transforming multi-dimensional evaluation indicators into a comprehensive objective function, with the optimization objective being to minimize the total cost. The constraint modeling unit constructs the policy constraints of the data center, including computing resource constraints, power capacity constraints, and other constraints. The optimization algorithm solution unit uses a genetic algorithm to iteratively update all candidate scheduling scheme combinations, calculates the comprehensive objective function value, and converges to obtain a set of optimal computing power task scheduling schemes. The strategy output unit outputs the optimal computing power task scheduling scheme, which is decomposed into an executable computing power scheduling instruction set and a power control instruction set, and then sent to the computing power scheduling engine and power control engine of the data center, respectively.

Images