Optimal configuration method of hybrid energy storage for data center considering life cycle carbon emissions

Abstract

The application discloses a data center hybrid energy storage optimization configuration method considering whole life cycle carbon emission, which can comprehensively consider the coupling relationship among the thermodynamic environment of the data center, the battery life attenuation and the carbon footprint. The method firstly constructs a dynamic PUE model based on thermodynamics and a batch processing load uniform matrix model; secondly, an active frequency division mechanism of the hybrid energy storage system is introduced, and a super capacitor is used to smooth high-frequency fluctuations to prolong the service life of a lithium battery; then, a dynamic implicit carbon emission model is established to quantify the influence of equipment attenuation on the whole life cycle carbon footprint. On this basis, a double-layer multi-objective optimization framework is proposed, the upper layer optimizes the capacity configuration through an NSGA-II algorithm, and the lower layer solves multiple typical days rolling scheduling through a CPLEX solver. In this way, not only the decoupling optimization of economic cost and carbon emission is realized, but also the service life and environmental benefits of the energy storage system are effectively improved, and the method is suitable for low-carbon planning and operation of green data centers.

Description

Technical Field

[0001] This application relates to the field of data center energy management technology, and in particular to a method for optimizing the configuration of hybrid energy storage in data centers that takes into account carbon emissions throughout the entire life cycle. Background Technology

[0002] With the development of the digital economy, data center energy consumption has risen sharply, making the integration of renewable energy and energy storage systems crucial for achieving "dual carbon" goals. However, existing technologies for data center energy storage configurations have the following shortcomings: First, traditional power efficiency estimates are mostly static values, ignoring the dynamic characteristics of the thermal coupling between ambient temperature and server load; second, single-cell battery energy storage struggles to meet the demands of both high energy density and high power density, and frequent power fluctuations accelerate battery aging; third, existing carbon emission calculations often ignore the implicit carbon from equipment production and disposal, and fail to consider the impact of battery lifespan degradation on dynamic implicit carbon; finally, traditional single-layer optimization models struggle to handle the complex coupling between capacity planning (long-cycle LCA and cost) and operational scheduling (short-cycle second / minute-level response), causing configuration schemes to deviate from the optimal solution in actual operation. Therefore, there is an urgent need for an optimized configuration method that can comprehensively consider the thermodynamic environment, battery life, and carbon emissions throughout the entire lifecycle. Summary of the Invention

[0003] This invention provides a method for optimizing the configuration of hybrid energy storage in data centers that considers carbon emissions throughout the entire life cycle. The method includes: constructing a refined load and energy consumption model for the data center. The model includes: a dynamic power utilization efficiency model based on environmental thermal balance, and a unified batch processing load matrix model based on underlying directed acyclic graph logic. The unified batch processing load matrix model is used to describe the arrival, execution status, and delay scheduling decisions of batch processing tasks.

[0004] A physical constraint and active frequency division model for a hybrid energy storage system is constructed, decoupling the hybrid energy storage system into an energy-type subsystem and a power-type subsystem. The power ramp-up rate of the energy-type subsystem is limited by the low-pass filtering principle, and the high-frequency power fluctuations are handled by the power-type subsystem.

[0005] An annualized evaluation strategy based on clustering of multiple typical daily scenarios is constructed. Historical data is clustered to obtain multiple typical scenarios and their frequency of occurrence. The optimal scheduling strategy under each typical scenario is solved in parallel, and the annual technical and economic indicators are estimated based on frequency weighting.

[0006] A full life-cycle carbon emission assessment model is constructed, which includes the carbon emissions embodied in the equipment and the carbon emissions from electricity during the operation phase. The carbon emissions embodied in the equipment include the initial configuration carbon emissions of the energy-type subsystem, the carbon emissions from battery replacement, and the carbon emissions embodied in the power-type subsystem. The carbon emissions from battery replacement are dynamically calculated based on the expected lifespan of the battery.

[0007] Construct and solve a two-level multi-objective optimization model, including:

[0008] The upper-level model uses the rated power and rated capacity of the energy storage system as decision variables, and the total life cycle cost and total life cycle carbon emissions as optimization objectives, and uses the NSGA-II algorithm to optimize capacity configuration.

[0009] The lower-level model uses the grid power purchase, the power of each energy storage subsystem, and the batch processing task scheduling decision as variables, and aims to minimize the comprehensive operating cost under typical scenarios. It is solved using the CPLEX solver. The daily battery throughput obtained from the lower-level solution is fed back to the upper level to correct the expected battery lifespan, thus forming a closed-loop coupling between capacity configuration and operation scheduling.

[0010] In some possible implementations, the dynamic power utilization efficiency model based on environmental thermal balance calculates the energy consumption of the refrigeration system based on the thermal balance equation, expressed as:

[0011] ;

[0012] in, For a moment Energy consumption of the refrigeration system For a moment Total energy consumption of the server cluster The heat transfer coefficient is... For the area of ​​the enclosure structure, For real-time energy efficiency ratio, Outdoor temperature Indoor temperature;

[0013] The calculation model for the total energy consumption of a server cluster is as follows:

[0014] ;

[0015] in, For a moment Total energy consumption of the server cluster For server type indexing, For the first Server-like idle power consumption For the first Peak power consumption under full load for server-like systems For a moment No. CPU utilization of the server For a moment The first one in running state Number of class servers For the first Server-like sleep power consumption For a moment The first one in hibernation state Number of class servers;

[0016] The server state conservation constraint is represented as follows:

[0017] ;

[0018] in, For the first Total number of servers of this type.

[0019] Among some possible implementations, a unified batch processing load matrix model based on underlying directed acyclic graph logic is constructed, including:

[0020] Define batch jobs The task set is used to establish three core matrices: execution status matrix, arrival rate matrix, and transition decision matrix.

[0021] The three types of matrices satisfy the following relationship:

[0022] ;

[0023] in, For convolution operators, To execute the state matrix, , Representing the Batch processing tasks, Represents a point in time when the system is running. Indicates task At any moment Execution in progress. Indicates task At any moment Not executed; For the arrival rate matrix, , Indicates task At any moment Arrived at the system; For the transition decision matrix, , For the delay duration, the element is... Indicates the decision to carry out the mission put off indivual Execute if successful, otherwise return 0;

[0024] Construct a satisfaction decay matrix based on job delay tolerance , , Indicates the task Delayed to time The corresponding penalty weights are applied, and different weight growth rules are set based on task priority; among them, the weight of ordinary tasks increases linearly with the delay time, while the weight of urgent tasks increases exponentially with the delay time.

[0025] In some possible implementations, the energy-type subsystem is a lithium battery, and the power-type subsystem is a supercapacitor; the specific process of constructing the physical constraints and active frequency division model of the hybrid energy storage system includes:

[0026] Setting lithium battery ramp-up and relaxation constraints is represented as follows:

[0027] ;

[0028] in, The filtering smoothing coefficient is... and for Time and How much power should the battery output at time -1? The rated capacity of the lithium battery, For the slack variables of the lithium battery ramping constraint;

[0029] The high-frequency power component allocation of supercapacitors based on power balance relationships is expressed by the following formula:

[0030] ;

[0031] ;

[0032] ;

[0033] in, For supercapacitors in The power it bears at all times For hybrid energy storage The power it bears at all times The power required by the load. for The amount of electricity purchased at any given time;

[0034] An iterative equation for the state of charge (SOC) of a supercapacitor is constructed to track the remaining charge of the supercapacitor in real time. The formula is expressed as follows:

[0035] ;

[0036] in, and This represents the remaining percentage of charge in the supercapacitor at the current and previous moments. For scheduling time intervals, The charging and discharging efficiency of supercapacitors. Configure the given supercapacitor rated capacity by the upper-level optimization model;

[0037] Set the SOC operating constraints for the supercapacitor as follows:

[0038] ;

[0039] in, and To determine the minimum remaining charge percentage that a supercapacitor can achieve. and It is a slack variable, for The overflow amount of the supercapacitor's SOC upper and lower bounds at any given time.

[0040] Among some possible implementations, annualized evaluation strategies based on clustering of multiple typical daily scenarios include:

[0041] For the same set of energy storage configuration schemes given by the upper layer, the optimal scheduling strategy under K typical scenarios is solved in parallel by the lower layer model. The daily operating cost and daily battery throughput of each typical scenario are obtained respectively. The formula for calculating the daily battery throughput is as follows:

[0042] ;

[0043] in, For the daily battery throughput, For the first In each scenario Battery power at any time, For scheduling step size;

[0044] Based on the frequency of occurrence of each typical scenario, the annual operating electricity cost, annual indirect carbon emissions, annual equivalent battery throughput, and expected battery life are weighted and estimated. The estimated annual technical and economic indicators are then input into the upper-level model for the calculation of the upper-level objective function.

[0045] In some possible implementations, the calculation process of the life-cycle carbon emission assessment model includes:

[0046] Calculating the hidden carbon emissions of supercapacitors The formula is expressed as:

[0047] ;

[0048] in, and The rated power and capacity of the supercapacitor are provided for optimization of the upper layer. The carbon emission factor per unit capacity of supercapacitors is determined by the production and transportation process. This is the capacity regularization coefficient;

[0049] Calculate the total implicit carbon emissions over the entire life cycle of lithium batteries The formula is expressed as:

[0050] ;

[0051] in, The carbon emission factor per unit capacity of lithium batteries is determined by the production and transportation process. To improve material recycling efficiency, Carbon emission reduction factors resulting from battery recycling Number of battery replacements;

[0052] The indirect carbon emissions from electricity during the entire lifecycle of a data center project are calculated and represented as follows:

[0053] ;

[0054] in, Scenes derived from optimization of the lower layer Down The actual power that the data center purchases from the power grid at any given time. The real-time average carbon emission factor of the power grid;

[0055] The total carbon emissions over the entire life cycle are the sum of the carbon emissions implicit in supercapacitors, the total implicit carbon emissions of lithium batteries, and the indirect carbon emissions from electricity during the operation phase.

[0056] In some possible implementations, the objective function of the upper-level model includes the total lifecycle cost. and total carbon emissions throughout the entire life cycle ;

[0057] Among them, total life cycle cost This includes initial investment cost, battery replacement cost, total lifecycle operating electricity cost, demand electricity cost, and depreciation cost based on battery lifespan.

[0058] The constraints of the upper-level model include: energy storage configuration capacity boundary constraints, which limit the rated power and rated capacity of lithium batteries and supercapacitors to within a reasonable range of preset equipment specifications; energy-power rate constraints, which limit the charge and discharge rates of lithium batteries to within a preset range; and feasibility constraints of the lower-level solution. If the lower-level CPLEX solver returns an infeasible solution or exceeds a preset threshold penalty cost for the energy storage configuration scheme given by the upper level, the fitness of the individual corresponding to that configuration scheme is set to the range, so that it is eliminated in the evolution process of the NSGA-II algorithm.

[0059] In some possible implementations, the decision variables of the lower-level model include: grid power purchase, lithium battery charging and discharging power, supercapacitor charging and discharging power, batch processing task transfer decision matrix, and time step. The number of various types of servers in operation, lithium battery ramp-up relaxation variables, and supercapacitor SOC upper and lower bound relaxation variables;

[0060] The optimization objective function of the lower-level model is to minimize the comprehensive operating cost of the data center on a typical single day. The comprehensive operating cost includes the time-of-use grid electricity purchase cost, the maximum daily demand electricity cost, the operating carbon tax cost, the battery life loss depreciation cost, the physical boundary violation penalty cost, and the user experience loss cost.

[0061] In some possible implementations, the constraints of the lower-level model include: system power balance constraints, transformer capacity hard constraints, hybrid energy storage frequency division and physical operation constraints, batch processing task flow conservation and time window constraints, DAG task dependency and timing constraints, batch processing load superposition peak constraints, and server dynamic resource supply and demand constraints.

[0062] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0063] In this invention, the traditional static PUE estimation is abandoned. A dynamic energy consumption model based on thermodynamics is established to accurately capture the nonlinear impact of ambient temperature and IT load on cooling energy consumption, improving the model's accuracy in reproducing real-world operating scenarios. This invention introduces an active frequency division mechanism from a hybrid energy storage system, utilizing supercapacitors to handle high-frequency power fluctuations, significantly reducing the micro-cycle losses of lithium batteries. Although this increases the initial carbon input of the supercapacitor, it significantly extends the lifespan of the lithium battery, reducing the number of battery replacements and achieving a net benefit in total carbon emissions over the entire lifecycle. In the two-layer optimization model, the upper-layer objective function deliberately removes carbon tax costs, eliminating the strong coupling between economic and environmental costs. This allows for searching a wider range of non-dominated solutions, providing comprehensive decision support for data center operators with different preferences, from extreme economics to extreme low carbon. The lower-layer scheduling model introduces a physical out-of-bounds penalty mechanism and relaxation variables, ensuring that even in extreme conditions such as insufficient supercapacitor capacity, the algorithm can still find physically feasible solutions by relaxing battery constraints, avoiding solution failures caused by rigid constraints in traditional algorithms. Attached Figure Description

[0064] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 A schematic flowchart of an embodiment of a data center hybrid energy storage optimization configuration method considering carbon emissions throughout its entire life cycle, provided by an embodiment of the present invention;

[0066] Figure 2 A flowchart of the method provided in an embodiment of the present invention;

[0067] Figure 3 This is a schematic diagram of the summer optimal scheduling in an embodiment of the present invention;

[0068] Figure 4 This is a schematic diagram of the optimal scheduling for spring and autumn seasons in an embodiment of the present invention;

[0069] Figure 5 This is a schematic diagram of the optimal scheduling in winter according to an embodiment of the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0071] In the relevant descriptions of this embodiment, the terms "including," "containing," and "possessing" are all open terms and are generally understood to include but not be limited to; the term "at least one" is generally understood to mean one or more, where "multiple" refers to two or more; the term "at least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items, for example, "at least one of a, b, or c", or "at least one of a, b, and c", which can all mean: a, b, c, ab (i.e., a and b), ac, bc, or abc, where a, b, and c can be single or multiple; the symbol "A / B" is used to describe the selection relationship of associated objects, generally indicating an "or" relationship.

[0072] In the following description of the embodiments, the terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms "a" and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0073] Those skilled in the art should understand that, in the following description of the embodiments of this application, the sequence of numbers does not imply the order of execution. Some or all steps may be executed in parallel or sequentially. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0074] Those skilled in the art will understand that the numerical ranges in the embodiments of this application should be understood to specifically disclose each intermediate value between the upper and lower limits of the range. Any stated value or intermediate value within a stated range, as well as any other stated value or each smaller range between intermediate values ​​within a range, are also included within this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0075] Unless otherwise stated, the technical / scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. While this application describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this application. All references to this specification are incorporated by way of citation to disclose and describe the methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0076] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0077] With the development of the digital economy, data center energy consumption has risen sharply, making the integration of renewable energy and energy storage systems crucial for achieving "dual carbon" goals. However, existing technologies for data center energy storage configurations have the following shortcomings: First, traditional power efficiency estimates are mostly static values, ignoring the dynamic characteristics of the thermal coupling between ambient temperature and server load; second, single-cell battery energy storage struggles to meet the demands of both high energy density and high power density, and frequent power fluctuations accelerate battery aging; third, existing carbon emission calculations often ignore the implicit carbon from equipment production and disposal, and fail to consider the impact of battery lifespan degradation on dynamic implicit carbon; finally, traditional single-layer optimization models struggle to handle the complex coupling between capacity planning (long-cycle LCA and cost) and operational scheduling (short-cycle second / minute-level response), causing configuration schemes to deviate from the optimal solution in actual operation. Therefore, a two-layer optimization configuration method that can comprehensively consider thermodynamic environment, battery life, and carbon emissions throughout the entire lifecycle is urgently needed.

[0078] To address the aforementioned problems, this invention proposes a hybrid energy storage optimization configuration method for data centers that considers carbon emissions throughout the entire lifecycle. This method constructs a dynamic energy consumption model that considers environmental thermal balance, introduces an active frequency division mechanism for hybrid energy storage, establishes a dynamic implicit carbon emission model coupled with cycle life, and employs a two-layer optimization algorithm nested with NSGA-II and CPLEX. This method can accurately calculate the actual degradation path and the carbon footprint throughout the entire lifecycle of the battery. Through closed-loop feedback between upper-layer capacity collaborative configuration and lower-layer collaborative operation scheduling, it achieves the dual minimization of data center operating costs and carbon emissions.

[0079] Figure 1This is a schematic flowchart of an embodiment of a data center hybrid energy storage optimization configuration method considering carbon emissions throughout its entire life cycle, provided by an embodiment of the present invention. (See attached diagram.) Figure 1 As shown, the above method may include:

[0080] S101, construct a refined load and energy consumption model for data centers. The model includes: a dynamic power usage efficiency (PUE) model based on environmental thermal balance, and a unified matrix model for batch processing load based on underlying directed acyclic graph (DAG) logic.

[0081] Among them, the unified matrix model for batch processing load is used to describe the arrival, execution status and delay scheduling decisions of batch processing tasks;

[0082] In some embodiments, the specific process of constructing a dynamic power utilization efficiency model based on environmental thermal balance includes:

[0083] Calculation time Total power consumption of the server cluster , represented as:

[0084] ;

[0085] in, For a moment Total energy consumption of the server cluster Indexes for server types, such as compute-based and storage-based. For the first Server-like idle power consumption For the first Peak power consumption under full load for server-like systems For a moment No. CPU utilization of the server For a moment The first one in running state Number of class servers For the first Server-like sleep power consumption For a moment The first one in hibernation state Number of class servers;

[0086] The server state conservation constraint is represented as follows:

[0087] ;

[0088] in, For the first Total number of servers of this type;

[0089] A dynamic power efficiency model based on environmental thermal balance calculates the energy consumption of the refrigeration system using the thermal balance equation. Specifically, it obtains the real-time outdoor temperature through an environmental temperature sensor. With indoor temperature Calculate the time based on the heat balance equation Refrigeration system energy consumption , represented as:

[0090] ;

[0091] in, The heat transfer coefficient is determined by the physical properties of building materials, such as the thickness and material of concrete and insulation layers. The area of ​​the enclosing structure is the total area of ​​the walls that connect the computer room to the outside world. For real-time energy efficiency ratio, the real-time outdoor temperature is obtained through an ambient temperature sensor, based on performance curves provided by the equipment manufacturer or fitting functions based on historical data. Find the energy efficiency ratio under the current operating conditions;

[0092] A dynamic power efficiency model based on environmental thermal balance is constructed based on the total energy consumption of server clusters and the energy consumption of cooling systems to realize the coupled calculation of data center energy consumption with environmental temperature and IT load.

[0093] In some embodiments, to achieve fine-grained control over periodic or sudden batch processing jobs, batch processing jobs are defined. The task set is used to establish a general matrix model. Specifically, a unified matrix model for batch processing load is constructed based on the underlying directed acyclic graph logic, including:

[0094] Define batch jobs The task set is used to establish three core matrices: execution status matrix, arrival rate matrix, and transition decision matrix.

[0095] Execution state matrix This indicates the actual points in time when a batch processing job consumed computing resources under the final scheduling scheme, represented as:

[0096] ;

[0097] in, Representing the Batch processing tasks, Represents a point in time when the system is running. Indicates task At any moment Execution in progress. Indicates task At any moment Not executed, including pending or completed.

[0098] Arrival rate matrix This represents the initial number of batch processing jobs entering the system at each point in time, expressed as:

[0099] ;

[0100] in, Indicates task At any moment We have arrived at the system.

[0101] Transition Decision Matrix This is the core decision variable that the optimization algorithm (CPLEX) needs to solve for. It determines how long to postpone the execution of the task, and is expressed as:

[0102] ;

[0103] in, For the delay duration, the element is... Indicates the decision to carry out the mission put off indivual Execute, otherwise return 0.

[0104] The three types of matrices satisfy the following relationship:

[0105] ;

[0106] in, It is a convolution operator;

[0107] Construct a satisfaction decay matrix based on job delay tolerance , , Indicates the task Delayed to time The corresponding penalty weights are applied, with different weight growth rules set based on task priority; specifically, it's a pre-calculated two-dimensional "penalty price list" constructed based on the job's latency tolerance. This tells the optimizer: if the first... The task has been postponed to [time]. Enforcement requires determining the amount of the penalty (i.e., the weight).

[0108] Among them, the weight of ordinary tasks increases linearly with the delay time, while the weight of emergency tasks increases exponentially with the delay time.

[0109] For example, a typical task (linear growth), such as setting element values: A penalty of 1 yuan is imposed for each hour of delay. This type of task is not sensitive to delay, and the optimizer will tend to postpone it until midnight when electricity prices are lowest. Urgent tasks (exponential growth), such as setting element values: As delays increase, penalties will skyrocket, and the optimizer will tend to schedule it as early as possible.

[0110] S102, construct a physical constraint and active frequency division model for the hybrid energy storage system, decouple the hybrid energy storage system into an energy-type subsystem and a power-type subsystem, use the low-pass filtering principle to limit the power ramp-up rate of the energy-type subsystem, and let the power-type subsystem bear the high-frequency power fluctuations;

[0111] In some embodiments, the energy-type subsystem is a lithium battery, and the power-type subsystem is a supercapacitor.

[0112] The specific process of constructing the physical constraints and active frequency division model of the hybrid energy storage system includes:

[0113] Setting up lithium battery ramp-up and relaxation constraints, utilizing the low-pass filtering principle to limit the lithium battery ramp-up rate, and performing filtering constraints and power allocation, is expressed as follows:

[0114] ;

[0115] in, The filtering smoothing coefficient is determined by the physical characteristics or filtering requirements of the battery manufacturer. High-frequency power fluctuations exceeding the ramp rate limit automatically flow into the supercapacitor. and for Time and The power output of the battery at time -1 is calculated by the optimization solver. The rated capacity of the lithium battery is given by the configuration of the upper-level optimization model. Relaxed variables are introduced to address the ramp-up constraints of lithium batteries and to prevent supercapacitors from exceeding their limits. Adjusting the lithium battery ramping constraints prioritizes the safety of the physical system; the optimization solver will automatically adjust this value. Within the physically feasible range, the objective function includes a penalty term. , It will remain at 0.

[0116] The high-frequency power component allocation of supercapacitors based on power balance relationships is expressed by the following formula:

[0117] ;

[0118] ;

[0119] ;

[0120] in, For supercapacitors in The power it bears at all times For hybrid energy storage The power it bears at all times The power required by the load. for The amount of electricity purchased at any given time;

[0121] An iterative equation for the state of charge (SOC) of a supercapacitor is constructed to track the remaining charge of the supercapacitor in real time. The formula is expressed as follows:

[0122] ;

[0123] in, and This represents the remaining percentage of charge in the supercapacitor at the current and previous moments. For scheduling time intervals, The charging and discharging efficiency of supercapacitors. Configure the given supercapacitor rated capacity by the upper-level optimization model;

[0124] Set the operating constraints for the supercapacitor's State of Charge (SOC) as follows:

[0125] ;

[0126] in, and To determine the minimum remaining charge percentage that a supercapacitor can achieve. and It is a slack variable, for The overflow amount of the supercapacitor's SOC upper and lower bounds at any given time.

[0127] S103. Construct an annualized evaluation strategy based on clustering of multiple typical daily scenarios. Cluster historical data to obtain multiple typical scenarios and their occurrence frequencies. Solve the optimal scheduling strategy under each typical scenario in parallel and calculate the annual technical and economic indicators based on frequency weighting.

[0128] In some embodiments, given the significant seasonality of data center load and environmental parameters, to eliminate evaluation bias caused by a single scenario, this embodiment of the invention introduces a weighted strategy based on K-means clustering:

[0129] Specifically, scene clustering is performed, selecting ambient temperature, light intensity, and basic IT load from historical years as feature vectors, and using the K-means algorithm to cluster the annual data into K typical scenes, and obtaining the occurrence frequency (days / years) of each scene.

[0130] Then, in the lower layer of the two-layer optimization, parallel solution and weighted projection are performed. For the same set of energy storage configuration schemes given in the upper layer, the optimal scheduling strategy under K typical scenarios is solved in parallel to obtain the daily operating cost and daily battery throughput for each scenario. The formula for calculating the daily battery throughput is as follows:

[0131] ;

[0132] in, For the first In each scenario Battery power at any time, For scheduling step size;

[0133] Finally, in the calculation of the objective function of the upper-level model (NSGA-II), based on the occurrence frequency of each typical scenario, K typical scenarios are selected to calculate the annual operating electricity cost, annual indirect carbon emissions, annual equivalent battery throughput and expected battery life in a weighted manner. The calculated annual technical and economic indicators are then input into the upper-level model for the calculation of the upper-level objective function.

[0134] S104, Construct a full life cycle carbon emission assessment model, which includes the carbon emissions hidden in the equipment and the carbon emissions from electricity during the operation phase;

[0135] Among them, the carbon emissions of equipment include the initial configuration carbon emissions of energy-type subsystems, the carbon emissions of battery replacement, and the carbon emissions of power-type subsystems. The carbon emissions of battery replacement are dynamically calculated based on the expected lifespan of the battery.

[0136] In some embodiments, the calculation process of the life-cycle carbon emission assessment model includes:

[0137] Calculating the hidden carbon emissions of supercapacitors Unlike lithium batteries, supercapacitors have an extremely long lifespan (>500,000 cycles), typically covering the entire project cycle without needing replacement. Therefore, it is a one-time investment. The implicit carbon emission formula for supercapacitors is expressed as:

[0138] ;

[0139] in, and The rated power and capacity of the supercapacitor are provided for optimization of the upper layer. The carbon emission factor per unit capacity of supercapacitors is determined by the production and transportation process. The capacity regularization coefficient is introduced in this invention because, in reality, the carbon emission contribution of supercapacitor power units is extremely low. Its function is to provide the optimizer with a weak non-zero gradient penalty, guiding the algorithm to automatically search for and converge to the minimum necessary capacity solution while satisfying the power smoothing constraint, thus balancing physical realism and algorithm robustness.

[0140] Calculate the total implicit carbon emissions over the entire life cycle of lithium batteries This includes the total hardware carbon emissions generated by the lithium battery throughout the project lifecycle, which consists of initial configuration carbon emissions and battery replacement carbon emissions, expressed by the formula:

[0141] ;

[0142] in, The carbon emission factor per unit capacity of lithium batteries is determined by the production and transportation process. To improve material recycling efficiency, Carbon emission reduction factors resulting from battery recycling Number of battery replacements;

[0143] Among them, the number of battery replacements A dual-mode calculation strategy is adopted, using a continuous lifetime loss model during optimization iterations and integer logic for result output. The calculation formula is as follows:

[0144] ;

[0145] in, The design life of the data center project is (for example, 15 years in this invention). This refers to the expected lifespan of the battery.

[0146] Battery expected lifespan Based on the ampere-hour throughput fatigue model, a fluctuation penalty factor is introduced, and the calculation formula is as follows:

[0147] ;

[0148] in, The rated total throughput (kWh) over the battery's entire lifespan is usually given in the manufacturer's specifications (e.g., 3000 cycles). capacity); For annual equivalent throughput, This represents the total number of typical scenarios (typical days). For the first Frequency of a typical scenario in a year (days / year). Let K be the daily battery throughput under scenario k.

[0149] The indirect carbon emissions from electricity during the entire lifecycle of a data center project are calculated and represented as follows:

[0150] ;

[0151] in, Scenes derived from optimization of the lower layer Down The actual power that the data center purchases from the power grid at any given time. The real-time average carbon emission factor of the power grid;

[0152] In some embodiments, time-sharing fitting curves are used to simplify calculations. The valley period can be set to 0.3 kg CO2 / kWh, the normal period can be set to 0.5 kg CO2 / kWh, and the peak period can be set to 0.8 kg CO2 / kWh.

[0153] In this embodiment of the invention, the total carbon emissions over the entire life cycle are the sum of the implicit carbon emissions of supercapacitors, the total implicit carbon emissions of lithium batteries, and the indirect carbon emissions from electricity during the operation phase.

[0154] S105, Construct and solve a two-layer multi-objective optimization model, including:

[0155] S1051, the upper-level model takes the rated power and rated capacity of the energy storage system as decision variables, the total life cycle cost and total life cycle carbon emissions as optimization objectives, and uses the NSGA-II algorithm to optimize capacity configuration.

[0156] S1052, the lower-level model uses the grid power purchase, the power of each energy storage subsystem, and the batch processing task scheduling decision as variables, and aims to minimize the comprehensive operating cost under typical scenarios. It is solved using the CPLEX solver. The daily battery throughput obtained from the lower-level solution is fed back to the upper-level model to correct the expected battery lifespan, thus forming a closed-loop coupling between capacity configuration and operation scheduling.

[0157] In some embodiments, the upper-level model is a capacity coordination configuration model, and the decision variables of the upper-level model are represented as follows:

[0158] ;

[0159] in, This refers to the rated power of the lithium battery. This refers to the rated capacity of the lithium battery. This refers to the rated power of the supercapacitor. This refers to the rated capacity of the supercapacitor.

[0160] In some embodiments, the objective function of the upper-level model includes the total lifecycle cost. and total carbon emissions throughout the entire life cycle ;

[0161] Specifically, total lifecycle cost Including initial investment costs Battery replacement cost Electricity costs throughout the entire life cycle Electricity demand Battery life loss cost ;

[0162] Initial investment cost , represented as:

[0163] ;

[0164] in, Price per unit of battery power. Price per unit of battery capacity This refers to the unit price of supercapacitors;

[0165] Replacement cost , represented as:

[0166] ;

[0167] in, The preset annualized reduction factor for battery costs (e.g., 0.05) is intended to account for technological advancements, where batteries will typically be cheaper to purchase years from now than they are now.

[0168] Operating electricity cost , represented as:

[0169] ;

[0170] in, For the first Daily operating costs under a typical scenario For the first The daily carbon tax cost in a typical scenario For the first Maximum electricity cost in a typical scenario For the first The cost of battery life loss in a typical scenario is calculated by the lower-level solver under the optimal scheduling strategy. Demand-based electricity pricing; annual peak power generation of the entire grid. The results of lower-level scheduling are fed back to the upper level, which is in all A typical scenario (and all moments) In this context, the maximum power that a data center can draw from the power grid is:

[0171] .

[0172] In some embodiments, the constraints of the upper-level model include:

[0173] Energy storage configuration capacity boundary constraints limit the decision variables such as the rated power and rated capacity of lithium batteries and supercapacitors to within a reasonable range of the preset equipment (e.g., existing equipment on the market) specifications;

[0174] Energy-power rate constraints limit the charge and discharge rates of lithium batteries to a preset range; for example, the power and capacity of lithium batteries are not completely independent and must conform to electrochemical characteristics, so a battery with extremely small capacity but extremely high power cannot be configured.

[0175] The feasibility constraints of the lower-level solution require that the upper-level configuration scheme be able to be computed by the lower level. For example, if the battery capacity passed from the upper level is too small, causing the lower level to cut off the load or the system to crash, then the configuration is invalid. If the lower-level CPLEX solver returns an infeasible solution or a penalty cost exceeding a preset threshold for the energy storage configuration scheme given by the upper level, then the fitness of the individual corresponding to that configuration scheme is set to the range, so that it is eliminated in the evolution process of the NSGA-II algorithm.

[0176] In some embodiments, the lower-level model is a cooperative operation scheduling model, and the decision variables of the lower-level model include: power purchased by the power grid. Lithium battery charging and discharging power Supercapacitor charging and discharging power Off-site dispatch volume Batch processing task transfer decision matrix ,time Number of various types of servers in operation Lithium battery ramp-up relaxation variables Supercapacitor SOC upper limit overflow Supercapacitor SOC lower limit deficit .

[0177] The objective function is to minimize the overall operating cost of the data center in each typical scenario. , represented as:

[0178] ;

[0179] The time-of-use electricity purchase cost is expressed as:

[0180] ;

[0181] in, This is the local time-of-use electricity price curve, for example, 0.3 yuan / kWh during off-peak hours, 0.8 yuan / kWh during normal hours, and 1.2 yuan / kWh during peak hours. This is the scheduling step size.

[0182] The maximum demand electricity charge is expressed as:

[0183] ;

[0184] in, This is the factor used to convert the demand electricity price to a single day, for example, 1 yuan / kW·day.

[0185] The cost of implementing a carbon tax is expressed as:

[0186] ;

[0187] in, This refers to the price of carbon trading.

[0188] The cost of battery life depreciation is expressed as follows:

[0189] ;

[0190] in, The conversion factor (yuan / kWh) is calculated using the following formula:

[0191] ;

[0192] The physical boundary violation penalty cost is expressed as:

[0193] ;

[0194] in, To ensure absolute physical constraints, the voltage is set to 10,000,000 times the peak electricity price, so that the supercapacitor will never exceed the SOC limit. For safety constraints, the limit is set to 1000 times the peak electricity price. In an emergency, if the super capacity is full and the high-frequency power cannot handle it, the battery is allowed to temporarily exceed the limit in order to prevent the system from crashing.

[0195] The cost of sacrificing user experience is expressed as:

[0196] ;

[0197] in, is the unit satisfaction compensation coefficient, and Trace is the trace of the matrix.

[0198] In some embodiments, the constraints of the lower-level model include: system power balance constraints, transformer capacity hard constraints, hybrid energy storage frequency division and physical operation constraints, batch processing task flow conservation and time window constraints, DAG task dependency and timing constraints, batch processing load superposition peak constraints, and server dynamic resource supply and demand constraints.

[0199] Specifically, the system power balance constraint is expressed as:

[0200] ;

[0201] A hard constraint on transformer capacity is expressed as:

[0202] ;

[0203] in, Rated capacity of the data center transformer.

[0204] Hybrid energy storage frequency division and physical constraints are expressed as:

[0205] ;

[0206] ;

[0207] Batch processing task flow conservation and time window constraints are expressed as follows:

[0208] ;

[0209] Directed Acyclic Graph (DAG) dependencies and timing constraints ensure that tasks proceed in an orderly manner; a subsequent task cannot begin until the predecessor task is completed. This can be represented as:

[0210] ;

[0211] in, and To execute the state matrix The elements in the table represent the successor and predecessor tasks in... The instantaneous execution state at any given moment (0 or 1).

[0212] The peak constraint for batch processing load aggregation is expressed as:

[0213] ;

[0214] ;

[0215] ;

[0216] in, For at any time The total power consumed by all batch processing jobs in the "running" state within the data center. For a single task Operating power For the task CPU utilization during runtime; This represents the remaining server resources. This represents the total peak power of the currently running servers. For the first The number of servers currently running the class. This represents the real-time power consumed by the interactive load. Safety margin is reserved.

[0217] Server dynamic resource supply and demand constraints are represented as:

[0218] ;

[0219] in, This is the maximum utilization limit for servers set for security reasons.

[0220] In some embodiments, the solution of the two-layer multi-objective optimization model of the present invention uses NSGA-II to generate and evolve individuals with energy storage configuration in the outer layer, and CPLEX is used to solve the inner layer, and parallel computing is used to accelerate the evaluation of the fitness function.

[0221] Specifically, the NSGA-II algorithm steps include:

[0222] 1. Population Initialization. The algorithm first randomly generates an initial population A of size N. t Each individual represents a set of decision variables.

[0223] 2. Non-dominated ranking. The population is ranked according to the objective function value. If individual a is no worse than individual b in all objectives, and better than b in at least one objective, then a is said to dominate b. First level ( Includes all non-dominated individuals in the population; removes... Then, the remaining non-dominated individuals constitute the second layer ( ), and so on.

[0224] 3. Crowding Degree Calculation. To ensure a uniform distribution of solutions on the Pareto front, the algorithm calculates the crowding degree of each individual within the same level, reflecting the density of solutions surrounding that individual. At the same level, individuals with higher crowding degrees (i.e., solutions with fewer surrounding solutions) have a higher retention priority to maintain population diversity.

[0225] 4. Genetic manipulation. Parents are selected using a tournament selection method. When comparing two individuals, those with lower non-dominant rank are preferred; if the ranks are the same, those with higher crowding are selected. Crossover and mutation operations are performed on the selected parents to produce a progeny population B of size N. t .

[0226] 5. Elite Preservation Strategy. Preserve the best of the best from parent generation A. t With offspring B tMerge to form a composite population of size 2N. Rank them according to their non-dominant hierarchy. , ...; Individuals are selected sequentially to enter the next generation A t+1 When the total number of individuals in a certain layer exceeds N, the layer is sorted from largest to smallest based on the degree of crowding, and the individuals ranked at the top are selected to fill the N slots.

[0227] 6. Looping and Termination. Repeat the above steps until the maximum number of generations is reached. The final output is a set of non-dominated solutions, i.e., Pareto optimal solutions, for decision-makers to choose from based on their preferences.

[0228] Compared to existing technologies, this invention abandons traditional static PUE estimation and establishes a thermodynamic-based dynamic energy consumption model to accurately capture the nonlinear impact of ambient temperature and IT load on cooling energy consumption, improving the model's accuracy in reproducing real-world operating scenarios. This invention introduces an active frequency division mechanism from a hybrid energy storage system, utilizing supercapacitors to handle high-frequency power fluctuations, significantly reducing micro-cycle losses in lithium batteries. Although this increases the initial carbon input of the supercapacitor, it achieves a net benefit in total carbon emissions over the entire lifecycle by significantly extending lithium battery life and reducing battery replacement frequency. In the two-layer optimization model, the upper-layer objective function of this invention deliberately removes carbon tax costs, eliminating the strong coupling between economic and environmental costs, thereby enabling the search of a wider range of non-dominated solutions and providing comprehensive decision support for data center operators with different preferences, from "extreme economy" to "extreme low carbon." This invention introduces a physical out-of-bounds penalty mechanism and relaxation variables in the lower-layer scheduling model, ensuring that even under extreme conditions such as insufficient supercapacitor capacity, the algorithm can still find physically feasible solutions by relaxing battery constraints, avoiding solution failures caused by rigid constraints in traditional algorithms.

[0229] The process of steps S101 to S105 described above will be specifically explained below with a specific embodiment.

[0230] Taking a data center in a certain location as the research object, with a design life of 15 years, the data center is equipped with a transformer with a rated capacity of 2000kVA. The peak power consumption of a single server in the IT system is 0.5kW, and the idle power consumption is 0.2kW. The indoor target temperature is set at 22℃, and the energy efficiency ratio of the cooling system fluctuates dynamically with the outdoor temperature. Data on outdoor ambient temperature, light intensity, basic IT load, number of batch processing tasks arriving, and total energy consumption of the server cluster are taken as examples for three typical days (summer, spring / autumn, and winter) in the data center.

[0231] The basic parameters are shown in Table 1:

[0232] Table 1

[0233]

[0234] Figure 2 See the flowchart of the method provided in the embodiment of the present invention. Figure 2 As shown:

[0235] First, calculate the time. Total power consumption of the server cluster With time Refrigeration system energy consumption A unified matrix model for batch processing load is constructed based on the underlying DAG logic and passed to the lower layer as part of the constraints of the lower layer.

[0236] Based on the decision variables given by the upper layer, a physical constraint and active frequency division model of the hybrid energy storage system is constructed. The hybrid energy storage system is decoupled into two subsystems: energy type (lithium battery) and power type (supercapacitor). The ramp-up and relaxation constraints of lithium battery and the state of charge constraints of supercapacitor are passed to the lower layer as part of the constraints of the lower layer.

[0237] An annualized evaluation strategy based on clustering of multiple typical day scenarios is constructed. The operating costs of typical days under the optimal scheduling obtained from the lower-level optimization are weighted and extrapolated to the technical and economic indicators for the whole year and then passed on to the upper level.

[0238] Construct a life-cycle carbon emission (LCA) assessment model to calculate the implicit carbon emissions of supercapacitors. Calculate the total hardware carbon emissions generated by the lithium battery throughout the entire project lifecycle. The calculation of indirect carbon emissions from electricity purchases over 15 years of data center operation is used to determine the expected battery lifespan based on the annual equivalent battery throughput. The battery replacement count is obtained, and finally the implicit carbon emissions, operational carbon emissions, and battery replacement count are passed to the upper layer.

[0239] The optimal configuration scheme obtained is shown in Table 2:

[0240] Table 2

[0241]

[0242] The optimal scheduling for three typical days in summer, spring / autumn, and winter is obtained under the optimal solution, and the operational performance is as follows: Figures 3 to 5 As shown, where, Figure 3 This is a schematic diagram of the summer optimal scheduling in an embodiment of the present invention. Figure 4 This is a schematic diagram of the optimal scheduling for spring and autumn seasons in an embodiment of the present invention. Figure 5 This is a schematic diagram of the optimal scheduling in winter according to an embodiment of the present invention.

[0243] based on Figures 3 to 5As shown, the optimized power purchase results for three typical days reflect the balance between economy and security achieved by the two-layer optimization configuration scheme. The system demonstrates its economic efficiency, with power purchases largely kept within a limit, which helps reduce demand charges and avoid peak electricity prices. The most significant optimization effect is observed between 18:00 and 21:00, where power purchases decrease, indicating that the system prioritizes the use of lithium battery storage energy to support the load, successfully avoiding periods of high electricity prices. The system also demonstrates its stability; during the high load period from 8:00 to 18:00, the system, through the rapid response of supercapacitors, ensures a relatively stable power purchase curve under fluctuating loads, preventing transformer overload.

[0244] The coordinated response of the hybrid energy storage system on three typical days clearly demonstrates the protective effect of the active frequency division mechanism on the lifespan of lithium batteries. The supercapacitor undertakes high-frequency power compensation throughout the day, with its curve exhibiting extremely dense charge-discharge switching. Its main function is to smooth out power spikes caused by IT load fluctuations, avoiding frequent deep charge-discharge cycles for the lithium batteries. The lithium batteries, on the other hand, charge during periods of low electricity prices and discharge during peak periods, alleviating pressure and reducing operating costs through peak shaving and valley filling.

[0245] Load and environmental thermodynamic analysis on three typical days demonstrates the impact of ambient temperature on data center cooling energy consumption, confirming the dynamic PUE modeling logic in the patent. Most notably, on a typical summer day, the air conditioning cooling power (blue dashed line) exhibits a high degree of coupling with temperature and total IT load (light blue shading). During the afternoon's high-temperature period, cooling power consumption slightly increases due to the increased indoor-outdoor temperature difference. Simultaneously, the IT load shows two significant peaks in batch processing tasks around 10:00 and 15:00, corresponding to the DAG-based task scheduling mentioned in the patent.

[0246] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0247] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A method for optimizing the configuration of hybrid energy storage in data centers, considering carbon emissions throughout the entire lifecycle, characterized in that, include: A refined load and energy consumption model for data centers is constructed. The model includes: a dynamic power utilization efficiency model based on environmental thermal balance, and a unified batch processing load matrix model based on underlying directed acyclic graph logic. The unified batch processing load matrix model is used to describe the arrival, execution status, and delay scheduling decisions of batch processing tasks. A physical constraint and active frequency division model for a hybrid energy storage system is constructed, decoupling the hybrid energy storage system into an energy-type subsystem and a power-type subsystem. The power ramp-up rate of the energy-type subsystem is limited by the low-pass filtering principle, and the high-frequency power fluctuations are handled by the power-type subsystem. An annualized evaluation strategy based on clustering of multiple typical daily scenarios is constructed. Historical data is clustered to obtain multiple typical scenarios and their frequency of occurrence. The optimal scheduling strategy under each typical scenario is solved in parallel, and the annual technical and economic indicators are estimated based on frequency weighting. A full life-cycle carbon emission assessment model is constructed, which includes the carbon emissions embodied in the equipment and the carbon emissions from electricity during the operation phase. The carbon emissions embodied in the equipment include the initial configuration carbon emissions of the energy-type subsystem, the carbon emissions from battery replacement, and the carbon emissions embodied in the power-type subsystem. The carbon emissions from battery replacement are dynamically calculated based on the expected lifespan of the battery. Construct and solve a two-level multi-objective optimization model, including: The upper-level model uses the rated power and rated capacity of the energy storage system as decision variables, and the total life cycle cost and total life cycle carbon emissions as optimization objectives, and uses the NSGA-II algorithm to optimize capacity configuration. The lower-level model uses the grid power purchase, the power of each energy storage subsystem, and the batch processing task scheduling decision as variables, and aims to minimize the comprehensive operating cost under typical scenarios. It is solved using the CPLEX solver. The daily battery throughput obtained from the lower-level solution is fed back to the upper level to correct the expected battery lifespan, thus forming a closed-loop coupling between capacity configuration and operation scheduling.

2. The method according to claim 1, characterized in that, The dynamic power efficiency model based on environmental thermal balance calculates the energy consumption of the refrigeration system based on the thermal balance equation, and the formula is expressed as: ； in, For a moment Energy consumption of the refrigeration system For a moment Total energy consumption of the server cluster The heat transfer coefficient is... For the area of ​​the enclosure structure, For real-time energy efficiency ratio, Outdoor temperature Indoor temperature; The calculation model for the total energy consumption of the server cluster is as follows: ； in, For a moment Total energy consumption of the server cluster For server type indexing, For the first Server-like idle power consumption For the first Peak power consumption under full load for server-like systems For a moment No. CPU utilization of the server For a moment The first one in running state Number of class servers For the first Server-like sleep power consumption For a moment The first one in hibernation state Number of class servers; The server state conservation constraint is represented as follows: ； in, For the first Total number of servers of this type.

3. The method according to claim 2, characterized in that, Construct a unified batch processing load matrix model based on underlying directed acyclic graph logic, including: Define batch jobs The task set is used to establish three core matrices: execution status matrix, arrival rate matrix, and transition decision matrix. The three types of matrices satisfy the following relationship: ； in, For convolution operators, To execute the state matrix, , Representing the Batch processing tasks, Represents a point in time when the system is running. Indicates task At any moment Execution in progress. Indicates task At any moment Not executed; For the arrival rate matrix, , Indicates task At any moment Arrived at the system; For the transition decision matrix, , For the delay duration, the element is... Indicates the decision to carry out the mission put off indivual Execute if successful, otherwise return 0; Construct a satisfaction decay matrix based on job delay tolerance , , Indicates the task Delayed to time The corresponding penalty weights are applied, and different weight growth rules are set based on task priority; among them, the weight of ordinary tasks increases linearly with the delay time, while the weight of urgent tasks increases exponentially with the delay time.

4. The method according to claim 3, characterized in that, The energy-type subsystem is a lithium battery, and the power-type subsystem is a supercapacitor; the specific process of constructing the physical constraints and active frequency division model of the hybrid energy storage system includes: Setting lithium battery ramp-up and relaxation constraints is represented as follows: ； in, The filtering smoothing coefficient is... and for Time and How much power should the battery output at time -1? The rated capacity of the lithium battery, For the slack variables of the lithium battery ramping constraint; The high-frequency power component allocation of supercapacitors based on power balance relationships is expressed by the following formula: ； ； ； in, For supercapacitors in The power it bears at all times For hybrid energy storage The power it bears at all times The power required by the load. for The amount of electricity purchased at any given time; An iterative equation for the state of charge (SOC) of a supercapacitor is constructed to track the remaining charge of the supercapacitor in real time. The formula is expressed as follows: ； in, and This represents the remaining percentage of charge in the supercapacitor at the current and previous moments. For scheduling time intervals, The charging and discharging efficiency of supercapacitors. Configure the given supercapacitor rated capacity by the upper-level optimization model; Set the SOC operating constraints for the supercapacitor as follows: ； in, and To determine the minimum remaining charge percentage that a supercapacitor can achieve. and It is a slack variable, for The overflow amount of the supercapacitor's SOC upper and lower bounds at any given time.

5. The method according to claim 4, characterized in that, The annualized evaluation strategy based on clustering of multiple typical daily scenarios includes: For the same set of energy storage configuration schemes given by the upper layer, the optimal scheduling strategy under K typical scenarios is solved in parallel by the lower layer model. The daily operating cost and daily battery throughput of each typical scenario are obtained respectively. The formula for calculating the daily battery throughput is as follows: ； in, For the daily battery throughput, For the first In each scenario Battery power at any time, For scheduling step size; Based on the frequency of occurrence of each typical scenario, the annual operating electricity cost, annual indirect carbon emissions, annual equivalent battery throughput, and expected battery life are weighted and estimated. The estimated annual technical and economic indicators are then input into the upper-level model for the calculation of the upper-level objective function.

6. The method according to claim 5, characterized in that, The calculation process of the life-cycle carbon emission assessment model includes: Calculating the hidden carbon emissions of supercapacitors The formula is expressed as: ； in, and The rated power and capacity of the supercapacitor are provided for optimization of the upper layer. The carbon emission factor per unit capacity of supercapacitors is determined by the production and transportation process. This is the capacity regularization coefficient; Calculate the total implicit carbon emissions over the entire life cycle of lithium batteries The formula is expressed as: ； in, The carbon emission factor per unit capacity of lithium batteries is determined by the production and transportation process. To improve material recycling efficiency, Carbon emission reduction factors resulting from battery recycling Number of battery replacements; The indirect carbon emissions from electricity during the entire lifecycle of a data center project are calculated and represented as follows: ； in, Scenes derived from optimization of the lower layer Down The actual power that the data center purchases from the power grid at any given time. The real-time average carbon emission factor of the power grid. For the first The frequency of a typical scenario throughout the year Design lifespan for data center projects; The total carbon emissions over the entire life cycle are the sum of the carbon emissions implicit in supercapacitors, the total implicit carbon emissions of lithium batteries, and the indirect carbon emissions from electricity during the operation phase.

7. The method according to claim 6, characterized in that, The objective function of the upper-level model includes the total lifecycle cost. and total carbon emissions throughout the entire life cycle ; Among them, total life cycle cost This includes initial investment cost, battery replacement cost, total lifecycle operating electricity cost, demand electricity cost, and depreciation cost based on battery lifespan. The constraints of the upper-level model include: energy storage configuration capacity boundary constraints, which limit the rated power and rated capacity of lithium batteries and supercapacitors to within a reasonable range of preset equipment specifications; energy-power rate constraints, which limit the charge and discharge rate of lithium batteries to within a preset range; and feasibility constraints of the lower-level solution, which set the fitness of the individual corresponding to the configuration scheme to the range if the lower-level CPLEX solver returns an infeasible solution or exceeds a preset threshold penalty cost for the energy storage configuration scheme given by the upper level, so that it is eliminated in the evolution process of the NSGA-II algorithm.

8. The method according to claim 7, characterized in that, The decision variables of the lower-level model include: grid power purchase, lithium battery charging and discharging power, supercapacitor charging and discharging power, batch processing task transfer decision matrix, and time step. The number of various types of servers in operation, lithium battery ramp-up relaxation variables, and supercapacitor SOC upper and lower bound relaxation variables; The optimization objective function of the lower-level model is to minimize the comprehensive operating cost of the data center on a typical single day. The comprehensive operating cost includes the time-of-use grid electricity purchase cost, the maximum daily demand electricity cost, the operating carbon tax cost, the battery life loss conversion cost, the physical boundary penalty cost, and the user experience loss cost.

9. The method according to claim 8, characterized in that, The constraints of the lower-level model include: system power balance constraints, transformer capacity hard constraints, hybrid energy storage frequency division and physical operation constraints, batch processing task flow conservation and time window constraints, DAG task dependency and timing constraints, batch processing load superposition peak constraints, and server dynamic resource supply and demand constraints.

Images