Centralized backup power energy storage optimization configuration method for data center based on cluster sharing architecture

Abstract

The application discloses a centralized standby power energy storage optimization configuration method for a data center based on a cluster sharing architecture. The method comprises the following steps: collecting historical operation data; constructing a reliability evaluation module, which is used for obtaining the reliability values of various data centers through sequential Monte Carlo method for fault timing simulation; constructing a centralized standby power energy storage configuration optimization model according to the reliability evaluation module; taking the energy capacity and the power capacity as the energy storage configuration parameters, taking the reliability constraint and the energy storage configuration constraint as the constraint conditions, taking the minimization of the energy storage cost as the objective function, and taking the combination function of the reliability penalty function and the objective function as the fitness function; and solving the centralized standby power energy storage configuration optimization model to obtain the optimal energy storage configuration parameters. The application can solve the centralized standby power energy storage configuration problem of the data center cluster sharing under the reliability constraint.

Description

Technical Field

[0001] This invention belongs to the fields of data center power supply and distribution system optimization and power system planning and optimization, specifically involving a method for optimizing the configuration of centralized backup power storage for data centers based on a cluster shared architecture. Background Technology

[0002] Data centers, as the cornerstone of my country's digital economy, are crucial infrastructure supporting the vigorous development of next-generation digital technologies such as cloud computing and artificial intelligence. Driven by the AI ​​wave, my country's intelligent computing power will experience rapid growth, and intelligent computing centers, which support this power, will exhibit significant characteristics of high computing power and high power density. Against this backdrop, data centers need to continuously improve the capacity of UPS or backup power storage to ensure their power supply reliability. However, data centers still face a relatively low probability of direct power outages from the mains power they are connected to, resulting in a significant decrease in the utilization rate of uninterruptible power supplies (UPS) and energy storage configured for reliability, creating a contradiction between energy storage resources and reliability effectiveness.

[0003] In pursuit of lower network latency and reduced marginal costs in infrastructure construction, future data centers will exhibit a new trend of clustered development, placing higher demands on the power distribution network construction of data center clusters. In recent years, the rise of shared energy storage, a new energy storage model, has provided a new approach to resolving the contradiction between energy storage resources and reliability in data centers. Specifically, centralized backup energy storage can be configured in a shared model within data center clusters to provide power support for data center operation in the event of a fault, thereby effectively improving the utilization rate of backup energy storage.

[0004] However, there is currently no shared energy storage configuration method that can simultaneously guarantee reliability, low cost, and efficient utilization of energy storage resources. Summary of the Invention

[0005] This invention addresses the problem of centralized backup power and energy storage in data centers under reliability constraints by providing an optimized configuration method for centralized backup power and energy storage based on a cluster shared architecture. First, it proposes a configuration form for backup power and energy storage in data center clusters. Then, it constructs a fault management module for situations involving missing power supply priorities and multiple power supply paths. A sequential Monte Carlo method is used to quantitatively assess the reliability of data center power supply and distribution. Finally, a centralized backup power and energy storage configuration optimization model is constructed and the objective function is improved. A heuristic algorithm is used to solve for the optimal capacity configuration. This invention provides an optimized configuration method for backup power and energy storage systems in data center clusters under reliability constraints, applicable to scenarios with rapidly increasing computing power demands, ensuring reliable power supply and efficient utilization of energy storage resources.

[0006] The technical solution adopted in this invention is as follows:

[0007] I. An Optimization Configuration Method for Centralized Backup Power Storage in Data Centers Based on Cluster Shared Architecture

[0008] The aforementioned method for optimizing the configuration of centralized backup power storage in data centers includes the following steps:

[0009] Step S1: Determine the design scheme of the data center cluster based on the shared architecture and the centralized backup power storage distribution network, and collect historical or planned operation data of the target data center cluster system.

[0010] Specifically, the target data center cluster system includes a centralized backup power storage station, at least one power source, and at least one data center; the target data center cluster system adopts a cluster shared architecture: the centralized backup power storage station includes energy storage batteries and energy storage buses; each data center includes a power supply and distribution system, a computer room, and a DC bus; the power supply and distribution system of each data center uses DC to supply power to the computer room, and the DC bus of each data center is connected to the energy storage bus of the centralized backup power storage station through DC cables, and the energy storage bus of the centralized backup power storage station is connected to the energy storage batteries through DC cables.

[0011] Step S2: Build a fault management module.

[0012] Specifically, the fault management module includes a power supply path confirmation unit and a multi-entity power allocation unit. The power supply path confirmation unit is used to obtain the power supply path of the data center and is configured to determine the power supply path of each data center by first using the power supply, then using centralized energy storage, and finally using centralized energy storage backup energy. The multi-entity power allocation unit is used to obtain the power supply capacity of the data center and is configured to: for the target data center cluster system or electrical island, if the total maximum power supply capacity is greater than or equal to the total power demand of the data center, it is determined that the output is sufficient, and the power supply capacity of each data center is equal to the power demand; if the total maximum power supply capacity is less than the total power demand of the data center, it is determined that the output is insufficient, and the total maximum power supply capacity is divided according to the ratio of the power demand of each data center to obtain the power supply capacity of each data center.

[0013] Step S3: Construct a reliability assessment module based on the fault management module. The reliability assessment module is used to perform fault timing simulation. When a fault occurs, the fault management module is used to determine the power supply path and power supply of each data center, and finally obtain the reliability value of each data center.

[0014] Specifically, the reliability refers to a combination of one or more power distribution network reliability indicators.

[0015] Specifically, the energy storage configuration constraints include upper and lower limits for energy capacity, upper and lower limits for power capacity, and upper and lower limits for the ratio of energy capacity to power capacity.

[0016] Step S4: Construct a centralized backup power storage configuration optimization model based on the reliability assessment module. The centralized backup power storage configuration optimization model uses the energy capacity and power capacity of the centralized backup power storage power station as energy storage configuration parameters, reliability constraints and energy storage configuration constraints as constraints, and minimizing energy storage cost as the objective function. A combination of the reliability penalty function and the objective function is used as the fitness function.

[0017] Step S5: Based on the historical operating data, solve the centralized backup power storage configuration optimization model using the fitness function to obtain the optimal energy storage configuration parameters.

[0018] Preferably, in step S5, a heuristic algorithm is used for iterative solution. During the iterative solution process, for each set of energy storage configuration parameters, the reliability assessment module is used to perform fault timing simulation to obtain the reliability value of each data center. The reliability value of each data center and the energy storage cost corresponding to the energy storage configuration parameters are input into the fitness function to obtain the fitness value.

[0019] In the preferred embodiment described above, the reliability assessment module can be used to perform M fault timing simulations, and the average value of the M original reliability values ​​for each data center can be obtained.

[0020] In the preferred embodiment described above, each fault timing simulation includes: randomly generating a composite fault timing sequence of the system using the sequential Monte Carlo method based on the failure rate and repair rate of each component; then, according to the composite fault timing sequence of the system, under the centralized backup power storage configuration corresponding to the energy storage configuration parameters, and in combination with the typical power load curves of each data center and the typical output curves of each power source, performing fault timing simulation; and obtaining the original reliability value of each data center based on the timing simulation data.

[0021] In each simulation time step of the fault timing simulation, if a component is faulty or the backup power capacity of the centralized backup power storage station is less than the preset backup power capacity threshold, the fault management module is used to obtain the power supply path and power of each data center in the target data center cluster system through a two-stage calculation method, and runs to the next simulation time step according to the power supply path and power; otherwise, the system maintains the normal power supply path and power and runs to the next simulation time step.

[0022] The two-stage calculation method is specifically as follows:

[0023] Phase 1: The minimum power load of each data center is taken as the power demand; the power supply path confirmation unit obtains the temporary power supply path and power supply of each data center based on the power demand of each data center and the power supply capacity of each power supply; it is determined whether the power supply of each data center can meet the power demand: if the power demand of the data center is met, the normal power load of the data center is taken as the new power demand; otherwise, the data center is considered to be in a power outage.

[0024] Phase Two: Using the power supply path confirmation unit, the power supply path and power output of each data center are obtained based on the new power demand of each data center and the power supply capacity of each power source.

[0025] In the two-stage calculation method, the power supply path confirmation unit obtains the power supply path and power supply of each data center through the following process:

[0026] First, the power supply is used as the power source to supply power to each data center according to the first power supply path. After using the connectivity algorithm to determine whether there is an electrical island in the target data center cluster system without considering centralized backup power storage facilities, for the target data center cluster system (when there is no electrical island) or each electrical island (when there is an electrical island), the multi-main power allocation unit is used to calculate the first power supply power of each data center according to the power demand of each data center, and obtain the remaining power demand of each data center.

[0027] The first power supply path is: power source - data center power distribution system - computer room;

[0028] Secondly, the power supply and centralized backup energy storage station are used as power supply sources to supply power to the data centers whose power demand is not met according to the second power supply path and the third power supply path, respectively. Using the multi-entity power distribution unit, the second power supply power of each data center whose power demand is not met is calculated according to the remaining power demand of each data center, and the load shedding of each data center is calculated.

[0029] The second power supply path is: power source - redundant power distribution system - energy storage bus - computer room;

[0030] The third power supply path is: energy storage battery - energy storage bus - computer room;

[0031] Finally, the power supply is used as the power source to charge the centralized backup energy storage station according to the fourth power supply path to obtain the energy storage battery charging power.

[0032] The fourth power supply path is: power source - redundant power distribution system - energy storage bus - energy storage battery.

[0033] The beneficial effects of this invention are as follows:

[0034] (1) This invention considers the characteristics of multi-data center clusters and provides a fault management module suitable for situations where power supply priorities are missing for multiple entities and loops exist in the network structure. This module is used to determine the power supply relationship between the data center, the mains power, and the centralized backup energy storage after a fault. Based on this, a reliability assessment module is proposed to evaluate the power supply reliability of data center clusters under centralized backup power, based on the sequential Monte Carlo method.

[0035] (2) Based on the shared centralized backup energy storage configuration problem considering low configuration cost and reliability constraints, this invention constructs a centralized backup energy storage configuration optimization model and applies a heuristic algorithm to solve it to obtain the optimal configuration scheme.

[0036] (3) This invention provides an innovative approach to the configuration of backup power and energy storage for data center clusters under reliability constraints, and has reference value for promoting the lean and scientific planning and design of power supply and distribution systems for computing facilities. Attached Figure Description

[0037] Figure 1 This is a flowchart of the overall process of the method of the present invention.

[0038] Figure 2 This is a diagram illustrating the configuration of a centralized energy storage power station in a DC-connected data center cluster according to the present invention.

[0039] Figure 3 This is a schematic diagram of the power supply path confirmation unit in this invention, showing the three steps of the power supply path confirmation process.

[0040] Figure 4 This is a flowchart (a) of the reliability assessment module in this invention and a diagram (b) of the fault timing generation and simulation.

[0041] Figure 5 This is a diagram of the data center cluster power distribution network system input in this embodiment of the invention;

[0042] Figure 6 This is a typical photovoltaic power output curve set in the embodiments of the present invention;

[0043] Figure 7 This is a typical wind power output curve diagram set in the embodiments of the present invention;

[0044] Figure 8 This is a typical power load demand diagram for a data center in an embodiment of the present invention;

[0045] Figure 9This is a graph showing the trend of reliability index changes under different shared centralized backup power storage configurations in the embodiments of the present invention; where (a) is the change of SAIFI index with capacity, (b) is the change of SAIDI index with capacity, (c) is the change of LOLE index with capacity, and (d) is the change of EENS index with capacity.

[0046] Figure 10 This is a scatter plot of the comprehensive energy storage configuration during the search process in an embodiment of the present invention.

[0047] Figure 11 This is a surface fitting diagram of the centralized energy storage integrated configuration cost according to an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0049] The first aspect of this invention provides a method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture.

[0050] like Figure 1 As shown, the data center centralized backup power storage optimization configuration method provided by the present invention specifically includes the following steps:

[0051] Step S1: Determine the design scheme of the data center cluster based on the shared architecture and the centralized backup power storage distribution network, and collect historical operating data of the target data center cluster system.

[0052] As a preferred embodiment of the present invention, the target data center cluster system includes a centralized backup power storage power station, at least one power source, and at least one data center.

[0053] In the preferred embodiment described above, the historical operating data of the target data center cluster system includes: the time-series power load curve of each data center, the time-series output curve of each power supply, and the failure rate and repair rate of each component.

[0054] As a preferred embodiment of the present invention, the target data center cluster system adopts a cluster sharing architecture.

[0055] like Figure 2As shown, the cluster shared architecture is as follows: the centralized backup power storage station includes energy storage batteries, energy storage converters (PCS), and energy storage buses; each data center includes a power supply and distribution system, a computer room, and a DC bus; the power supply and distribution system of each data center uses DC to supply power to the computer room, the power supply from the distribution network reaches the low-voltage AC bus via an AC transformer, and then the AC power is converted to DC power via an AC-DC converter and supplied to the computer room via the DC bus. The DC bus of each data center is connected to the energy storage bus of the centralized backup power storage station via DC cables, and the energy storage bus of the centralized backup power storage station is connected to the energy storage batteries via DC cables; during normal operation, the power supply path from any power source to any data center is uniquely determined.

[0056] The components include one or more electrical components such as lines, transformers, circuit breakers, and switches. The power source includes one or more of mains power, distributed generators, and renewable energy power plants.

[0057] Step S2: Construct a fault management module. The fault management module is used to determine the power supply path and power supply of each data center in the event of a fault, based on the load characteristics of the data center, in scenarios where power supply priority is missing or loops exist in the network structure. It includes a power supply path confirmation unit and a multi-entity power allocation unit. The power supply path confirmation unit is used to obtain the power supply path of the data center, and the multi-entity power allocation unit is used to obtain the power supply of the data center.

[0058] As a preferred embodiment of the present invention, the multi-entity power distribution unit is configured as follows: for the target data center cluster system or any electrical island within it, if the total maximum power supply capacity of the power supply is greater than or equal to the total power demand of the data center, then the power output is determined to be sufficient, and the power supply capacity of each data center is equal to the power demand; if the total maximum power supply capacity of the power supply is less than the total power demand of the data center, then the power output is determined to be insufficient, and the total maximum power supply capacity is divided according to the ratio of the power demand of each data center to obtain the power supply capacity of each data center.

[0059] Specifically: Assuming a power distribution network contains several data centers and power supplies, where the power supply path from any power supply j to any data center i is uniquely determined, the total maximum power supply capacity of the power supplies and the total power demand of the data centers can be calculated using the following formula:

[0060] P L sum =∑ i P L,i

[0061] P G sum =∑ j P G,j

[0062] In the formula, P G sum P represents the maximum power supply capacity of the power supply. L sum For the total power demand of the data center, P L,i To meet the power requirements of data center i, P G,j This represents the maximum power supply capacity of power source j.

[0063] For the two scenarios of sufficient output and insufficient output:

[0064] ①When P G sum ≥P L sum At that time, the power supply output can meet the load requirements of all data centers, and the power supply capacity of data center i is equal to the power demand P. L,i ;

[0065] ②When P G sum <P L sum When the power supply output is insufficient to meet the load demands of all data centers, the power supply capacity of each data center is calculated using the following formula:

[0066] P L,i actual =P G sum ·(P L,i / P L sum )

[0067] In the formula, P L,i actual For the power supply of data center i, P L,i To meet the power requirements of data center i, P L sum For the total power demand of the data center, P G sum This refers to the maximum power supply capacity of the power source.

[0068] As a preferred embodiment of the present invention, the power supply path confirmation unit is configured to: determine the power supply path to each data center based on the cluster shared architecture, according to the method of first using power supply, then using centralized energy storage, and finally centralized energy storage backup energy.

[0069] Specifically: such as Figure 3 As shown, when the power supply path confirmation unit is running, it performs the following process:

[0070] Step 1: Prioritize Power Supply: Use the power supply within the AC distribution network (mains power, diesel generators, etc.) as the power source, supplying power to the data center via the power supply path of "AC distribution network - data center power distribution system - computer room". Calculate the power supply capacity of each data center according to the multi-entity power distribution unit, and obtain the remaining power demand of each data center using the following formula:

[0071] P L,i,Step2 =P L,i,Step1 -P L,i,Step1 actual

[0072] In the formula, P L,i,Step1 To meet the power requirements of data center i, P L,i,Step1 actual P is the primary power supply for data center i. L,i,Step2 This represents the remaining power demand of data center i.

[0073] In the event of partial component failure (such as line interruption), a single distribution network may be split into multiple isolated networks. In this case, it is necessary to use a multi-main power distribution unit to identify the power supply path and power supply for each isolated network.

[0074] Step 2: Centralized Energy Storage Supplements Power Supply: Power is supplied to the data centers whose needs were not met in Step 1 via power supply paths provided by the AC distribution network (mains power, diesel generators, etc.) and the centralized backup energy storage station. These paths follow the power supply paths of "AC distribution network - redundant power distribution system - energy storage bus - data center" and "energy storage battery - energy storage bus - data center," respectively. The power supply capacity for each path is calculated using the multi-entity power distribution unit, and the load shedding capacity for data center i is determined using the following formula:

[0075] P L,i shedding =P L,i,Step2 -P L,i,Step2 actual

[0076] In the formula, P L,i,Step2 For the remaining power demand of data center i, P L,i,Step2 actual P is the second power supply for data center i. L,i shedding This is the load shedding amount of data center i.

[0077] Step 3: Maximize backup power: Use the power sources in the AC distribution network (mains power, diesel generators, etc.) as the power supply source, and charge the energy storage battery of the centralized backup energy storage station according to the power supply path of "AC distribution network - redundant power distribution system - energy storage bus - energy storage battery", and determine the charging power of the energy storage battery.

[0078] Step S3: Based on the fault management module, construct a reliability assessment module. The reliability assessment module is used to perform fault time-series simulation using the sequential Monte Carlo method. When a fault occurs, i.e., when a component is faulty or the backup power capacity of the centralized backup power storage station is less than the preset backup power capacity threshold, the fault management module is used to determine the power supply path and power supply of each data center. Finally, based on the time-series simulation data, the reliability value of each data center is obtained, thereby realizing the quantitative analysis of the reliability of each main body in the system.

[0079] In a preferred embodiment of the present invention, the reliability assessment module uses a fault management module to perform a two-stage calculation considering the data center load characteristics to obtain the optimal power supply path and power supply. The specific steps are as follows:

[0080] Phase 1: The load of critical data center equipment when it is powered on is taken as the minimum power load. Let the power demand of each data center be the minimum power load P. base That is, P L,i (t)=P base The power supply path confirmation unit confirms the power supply path and power output: ① If, at fault time t, the power demand P of data center i is... L,i (t)=P base If this condition can be met, then data center i can be considered to have uninterrupted power supply at that moment, allowing the power demand P for the second round of computation to be reduced. L,i (t) equals the normal power load of data center i; ② If the power load of data center i cannot be met... Electricity demand P L,i (t)=P base If data center i is considered to be experiencing a power outage, then the power demand P in the second phase will be reduced. L,i (t)=0.

[0081] Phase Two: Based on the electricity demand P obtained in Phase One L,i (t), using the power supply path confirmation unit, calculate the final power supply path and power supply.

[0082] As a preferred embodiment of the present invention, such as Figure 4 As shown, the fault timing simulation process is as follows:

[0083] Step S3.1) Initialize the simulation sequence number m=0, the simulation time t=0, the total number of simulations M, and the duration of a single simulation T.

[0084] Step S3.2) Assuming a component has a corresponding failure rate λ and a repair rate μ, calculate the component's fault-free operating time T using the following formula. TTF and fault repair time T TTRSum them up until the single simulation duration \(T\) is reached. By the above method, the fault time sequences of all components in the system can be obtained. The integration of the fault time sequences of all components can yield the composite fault time sequence of the system.

[0085] \(T\) TTF \(= -\ln(\theta_1) / \lambda\)

[0086] \(T\) TTR \(= -\ln(\theta_2) / \mu\)

[0087] [[ID=**13**]]Where the random numbers \(\theta_1, \theta_2 \in uniform(0,1)\) are used to simulate the random fault - free time and fault repair time of components.

[0088] Step S3.3) If there is a fault in the system or the backup power capacity \(SOE\) of the energy storage is lower than the set value \(SOE\) at a certain moment set , then based on the power output of the power source, the data center load, and the system topology at this moment, use the fault management module to perform two - stage calculations considering the load characteristics of the data center, and record indicators such as data center power outage and load shedding. Let \(t = t + 1\). If \(t < T\), repeat step S3.3; otherwise, jump to step S3.4.

[0089] Step S3.4) The single - fault time sequence simulation is completed. Statistically analyze the reliability indicators of the system running with the single - time sequence. Let \(m = m + 1\). If \(m < M\), then set \(t = 0\) and jump to step S3.2; otherwise, jump to step S3.5.

[0090] Step S3.5) After \(M\) times of fault time sequence simulations are completed, take the obtained average reliability value of the objective function as the reliability value.

[0091] Step S4. According to the reliability evaluation module, construct a centralized backup energy storage configuration optimization model; the centralized backup energy storage configuration optimization model takes the energy capacity and power capacity of the centralized backup energy storage power station as energy storage configuration parameters, takes reliability constraints and energy storage configuration constraints as constraint conditions, takes minimizing the energy storage cost as the objective function, adopts a penalty function mechanism, constructs a reliability penalty function according to the reliability constraints, and combines it with the objective function to obtain a fitness function.

[0092] Among them, the objective function can be expressed as:

[0093] \(\min f(E\) SESS , \(C\) SESS )=\(\sigma_1\cdot E\) SESS +\(\sigma_2\cdot C\) SESS

[0094] Where \(E\) SESS represents the energy capacity, \(C\) SESS represents the power capacity, \(\sigma_1\) represents the cost coefficient of the energy capacity, and \(\sigma_2\) represents the cost coefficient of the power capacity.

[0095] The reliability constraint refers to the requirement that the reliability values ​​of all data centers meet the preset reliability requirements. In a preferred embodiment of this invention, reliability is one or a combination of common power distribution network reliability indicators such as System Average Number of Outages (SAIFI), System Average Outage Time (SAIDI), Load Shedding Expectation (LOLE), and Energy Shortage Expectation (EENS).

[0096] Specifically, reliability constraints can be expressed as:

[0097] r i (E SESS C SESS )≤r i req

[0098] r∈R={SAIFI,SAIDI,LOLE,EENS}

[0099] In the formula, R is a set of reliability indicators, which can include various reliability indicators, and r∈R is a specific reliability indicator. i (E SESS C SESS ) represents the reliability level under the corresponding energy storage configuration, r i req This represents the preset reliability requirements for data center i.

[0100] Among them, the energy storage configuration constraints include the upper and lower limits of energy capacity, the upper and lower limits of power capacity, and the upper and lower limits of the ratio of energy capacity to power capacity.

[0101] Specifically, energy storage configuration constraints can be expressed as:

[0102] E SESS MIN ≤E SESS ≤E SESS MAX

[0103] C SESS MIN ≤C SESS ≤C SESS MAX

[0104] β1≤(E SESS / C SESS )≤β2

[0105] In the formula, E SESS MIN E SESS MAX These are the upper and lower limits of energy capacity, respectively; C SESSMIN C SESS MAX β1 and β2 are the upper and lower limits of power capacity, respectively; β1 and β2 are the upper and lower limits of the ratio of energy capacity to power capacity, respectively.

[0106] In step S4, the process of constructing a reliability penalty function based on reliability constraints using a penalty function mechanism includes: constructing a penalty function where the penalty factor is 1 when any data center meets its reliability requirements, and a significantly larger penalty factor when the reliability of any data center in the cluster does not meet its reliability requirements.

[0107] Preferably, the penalty function P(r) can be expressed by the following formula:

[0108]

[0109] In the formula, r∈R is a specific reliability index, r req λ represents the reliability requirement value, and λ is a large penalty coefficient.

[0110] In step S4, the combined function obtained by combining the reliability penalty function and the objective function serves as the fitness function and can be expressed as:

[0111]

[0112] In the formula, Cost(SESS) is the fitness function, and N IDC For the total number of data centers, SESS = (E SESS C SESS ) represents the energy storage configuration parameters, and f() is the objective function.

[0113] In step S4, the centralized backup power storage configuration optimization model can be expressed as:

[0114]

[0115] Step S5: Based on historical operating data, solve the centralized backup power storage configuration optimization model through the fitness function to obtain the optimal energy storage configuration parameters (optimal energy capacity and optimal power capacity) of the centralized backup power storage power station, which serves as the energy storage configuration result for the target data center cluster system.

[0116] In step S5, an iterative solution is performed using a heuristic algorithm. The heuristic algorithm mainly includes the following steps:

[0117] Step S5.1) Initialize N p There are k particles, and the initial position x is initialized for each particle k. k 0 With initial velocity vk 0 As shown in the following formula, let the iteration number it = 0:

[0118] x k 0 =SESS k 0 =(E SESS 0 C SESS 0 )

[0119] v k 0 =△SESS k 0 =(△E SESS 0 ,△C SESS 0 )

[0120] In the formula, SESS k 0 =(E SESS 0 C SESS 0 ) represents the initial centralized backup energy storage configuration for the k-th particle, ΔSESS k 0 =(△E SESS 0 ,△C SESS 0 ) represents the initial velocity of the k-th particle.

[0121] Step S5.2) In the it-th iteration, the reliability value of the centralized backup power storage configuration corresponding to each particle position is evaluated through the reliability assessment module, the fitness value is calculated according to the fitness function, and the optimal fitness position p is updated. best,k and the global optimal fitness position g best Substitute the values ​​into the following formula to update the particle's velocity and position.

[0122] v k it+1 =c0(it)·v k it +c1(it)·ρ1(p best,k -x k it )+c2(it)·ρ2(g best -x k it )

[0123]

[0124] In the formula, xk it It is the position of particle k in the it-th iteration, v k it Let be the velocity of particle k in the it-th iteration, c0(it) be the particle's inertia coefficient, c1(it) be the particle's self-learning coefficient, and c2(it) be the particle's social learning coefficient. The inertia coefficient, self-learning coefficient, and social learning coefficient can be configured as an expression related to the iteration number it to optimize the algorithm's convergence performance. ρ1, ρ2 ∈ uniform(0,1) are random numbers. The Bound() function is used to discretize the continuous centralized backup power storage capacity configuration. Let x be the position of particle k before the discretization of the centralized backup power storage capacity configuration in the (it+1)th iteration. k it+1 Let k be the discretized position of particle k after the centralized backup power storage capacity configuration in the (it+1)th iteration.

[0125] Step S5.3) Let the iteration number it = it + 1, if it <N iter If the search is successful, proceed to step S5.2; otherwise, end the search and set the current global optimal position g. best As the optimal centralized backup power storage configuration, the corresponding fitness Cost(g) best The configuration cost is considered as the optimal centralized backup power storage configuration.

[0126] As a preferred embodiment of the present invention, in step S5.2, for a set of energy storage configuration parameters corresponding to each particle, a fault timing simulation is performed using a reliability assessment module to obtain the reliability value of each data center; the reliability value of each data center and the energy storage cost corresponding to the energy storage configuration parameters are input into the fitness function to obtain the fitness value.

[0127] As a preferred embodiment of the present invention, the reliability assessment module is used to perform M failure timing simulations, and the average value of the M original reliability values ​​of each data center is taken to obtain the reliability value.

[0128] Each fault timing simulation includes: randomly generating a composite fault timing sequence of the system based on the failure rate and repair rate of each component; then using the sequential Monte Carlo method, and according to the composite fault timing sequence of the system, under the centralized backup power storage configuration corresponding to the energy storage configuration parameters, combined with the typical power load curves of each data center and the typical output curves of each power source, to perform fault timing simulation; and obtaining the original reliability value of each data center based on the timing simulation data.

[0129] As a preferred embodiment of the present invention, in each simulation time step of the fault timing simulation, if a component is faulty or the backup power capacity of the centralized backup power storage station is less than the preset backup power capacity threshold, the fault management module is used to obtain the power supply path and power supply of each data center in the target data center cluster system through a two-stage calculation method, and the system runs to the next simulation time step according to the obtained power supply path and power supply; otherwise, the system maintains the normal power supply path and power supply and runs to the next simulation time step.

[0130] The two-stage calculation method is as follows:

[0131] Phase 1: The minimum power load of each data center in the target data center cluster system or any isolated electrical network is taken as their respective power demand; the power supply path confirmation unit obtains the temporary power supply path and power supply capacity of each data center based on the power demand of each data center and the power supply capacity of each power source; it is determined whether the power supply capacity of each data center can meet the power demand: if the power demand of the data center is met, the original planned power load (normal power load) of the data center is taken as the new power demand; otherwise, the data center is regarded as a power outage, that is, 0 is taken as the new power demand.

[0132] The power supply capacity of each power source can be obtained from the time-series output curve, and the minimum power load of each data center can also be obtained from the time-series power load curve.

[0133] Phase Two: Using the power supply path confirmation unit, the power supply path and power output of each data center are obtained based on the new power demand of each data center and the power supply capacity of each power source.

[0134] In the calculations of the first and second phases, the power supply path confirmation unit obtains the power supply path and power output for each data center through the following process:

[0135] First, the power supply is used as the mains power source, supplying power to each data center according to the first power supply path. Using graph theory, it is determined whether electrical islands exist in the target data center cluster system (excluding centralized backup power storage facilities). Then, for the target data center cluster system (when no electrical islands exist) or each electrical island (when electrical islands exist), a multi-entity power distribution unit is used to calculate the first power supply to each data center based on its power demand and the maximum power supply capacity of the mains power source, thus obtaining the remaining power demand of each data center. The first power supply path is: mains power source - data center power distribution system - server room.

[0136] Secondly, for data centers with remaining power demand greater than zero (i.e., unmet power demand), the power supply and centralized backup energy storage station are used as power sources. Power is supplied to these data centers via the second and third power supply paths, respectively. Using a multi-entity power distribution unit, the second power supply capacity for each unmet data center is calculated based on its remaining power demand and the maximum power supply capacity of the power source. The load shedding for each data center is also calculated. The load shedding for the data center that ultimately meets its power demand can be set to zero. The second power supply path is: power supply - redundant power distribution system - energy storage bus - data center. The third power supply path is: energy storage battery - energy storage bus - data center.

[0137] Finally, the power source is used as the power supply to charge the centralized backup energy storage station according to the fourth power supply path, thus obtaining the charging power of the energy storage battery. The fourth power supply path is: power source - redundant power distribution system - energy storage bus - energy storage battery.

[0138] It should be noted that, under the first power supply path, the target data center cluster system that does not consider centralized backup power storage facilities means that, under the first power supply path, since shared energy storage power supply is not considered, the electrical islanding judgment under the first power supply path does not need to consider shared energy storage and related equipment. Therefore, after stripping the shared energy storage bus, DC cable, and energy storage station battery, the graph theory algorithm for judging connectivity is used to determine whether electrical islanding exists.

[0139] The second aspect of this invention provides a centralized backup power storage system for data centers based on a cluster-shared architecture.

[0140] The centralized backup power and energy storage system for data centers provided by this invention includes:

[0141] A centralized backup power and energy storage system for data centers includes at least one power source, at least one data center, and a centralized backup power and energy storage power station.

[0142] The centralized backup power storage system for data centers adopts a cluster-shared architecture. Specifically, the centralized backup power storage station includes energy storage batteries and energy storage buses; the data center includes a power supply and distribution system, a computer room, and a DC bus; the power supply and distribution system of each data center uses a DC scheme to power the computer room, and the DC bus of each data center is connected to the energy storage bus of the centralized backup power storage station through DC cables. The energy storage bus of the centralized backup power storage station is connected to the energy storage batteries through DC cables; the power supply path from any power source to any data center load is uniquely determined.

[0143] Specific embodiments of the present invention are as follows:

[0144] Example 1

[0145] This embodiment is an example of steps S1 to S3 in the method provided by the present invention, used to perform reliability assessment in a specific power distribution network system that includes power sources, multiple data centers and centralized backup energy storage power stations.

[0146] In step S1, the power distribution network system input in Example 1 is as follows: Figure 5 As shown, the power distribution network system consists of power sources (two municipal power sources and one distributed new energy power station), three data centers, and one centralized backup power storage power station (shared energy storage). Figure 5 In the diagram, T1~T5 represent transformers.

[0147] In step S1, the upper limit of mains power output in the distribution network system of Example 1 is set as the upper limit of power transmission of transformers T1 and T2 within the distribution network; such as Figure 6 and Figure 7 As shown, the power output curve of the new energy power plant is based on the annual wind and solar power output data provided by the ENTSO platform. After normalization and scenario reduction, the typical power output curve of new energy used in this embodiment is obtained; as shown Figure 8 As shown, typical power load demand curves for each data center within the power distribution network system are defined.

[0148] In step S1, based on empirical values, the failure rate and repair rate of the components in the power distribution network system are set as shown in Table 1 below.

[0149] Table 1 Failure rate and repair rate of different components

[0150] Component type Failure rate (times / year) Repair rate (times / hour) transformer 0.125 0.008 Overhead lines 0.200 0.083 Cable 0.200 0.055 disconnect switch 0.200 0.083 AC-DC converter 1.000 0.055 Mains electricity 0.400 0.200 New energy power station 1.000 0.083 Energy storage power station 1.000 0.167 busbar 0.300 0.200

[0151] In this embodiment, the system average number of outages (SAIFI), system average outage time (SAIDI), load shedding expectation (LOLE), and power shortage expectation (EENS) are selected as evaluation indicators for the reliability assessment method.

[0152] In steps S2 to S3, the key parameters used in the reliability assessment method of the power distribution network system are set as shown in Table 2.

[0153] Table 2 Key parameters used in the reliability assessment solution

[0154] parameter numerical values Simulation duration T (h) 8760 Number of simulations M 8192 Energy storage charge and discharge efficiency η 0.95 <![CDATA[Minimum power consumption P base (kW)]]> 500 <![CDATA[Energy storage set capacity SOE set (%)]]> 80

[0155] In this embodiment, with E fixed SESS And C SESS The ratio of E SESS / C SESS With the relationship of 5 unchanged, the centralized backup power storage capacity C SESS ini From the initial value C SESS ini=100kW with △C SESS =100kW step size gradually increases to C SESS fin =5MW, assess the power supply reliability of the data center cluster distribution network corresponding to each energy storage capacity.

[0156] By implementing the above method, different backup power storage capacities ΔC can be addressed. SESS The power supply and distribution reliability of the distribution network system with a capacity of 100kW was evaluated, and the curves of the power supply and distribution reliability index of each data center as a function of energy storage capacity were obtained, as shown in the figure. Figure 9 As shown.

[0157] Depend on Figure 9 It is evident that, on the one hand, with the increase in energy storage capacity, the power supply reliability of the data center cluster distribution network improves. Specifically, the SAIFI and SAIDI indicators show a more significant decrease, while LOLE and EENS decrease steadily, with the initial decrease occurring at a faster rate than the later one. This is in contrast to the situation without centralized energy storage configurations (E... SESS =0) and E SESS At a power consumption of 3000kW, the reliability indicators for LOLE and EENS decreased by 10.2% and 13.0% respectively, while SAIFI and SAIDI decreased by 88.9% and 97.0% respectively. This indicates that introducing shared centralized energy storage into the data center cluster distribution network can effectively improve power supply reliability. On the other hand, looking at the reliability indicators of different data centers, the reliability of Data Center 1 and Data Center 2 is similar, while the reliability of Data Center 3 is significantly worse. This is because the former is closer to the grid connection point of the distribution network, where grid reliability is high, while Data Center 3 is closer to a distributed renewable energy power station, which has significant volatility and high uncertainty.

[0158] Example 2

[0159] This embodiment is an example of steps S1 to S5 in the method provided by the present invention, used to optimize the configuration of centralized backup energy storage capacity in a specific power distribution network system that includes power sources, multiple data centers and centralized backup energy storage power stations.

[0160] In step S1, the power distribution network system input in Example 2 is as follows: Figure 5 As shown, the power distribution network system consists of two municipal power sources, one distributed new energy power station, three data centers (IDC1~IDC), and one shared energy storage system. Figure 5 In the diagram, T1~T5 represent transformers.

[0161] In step S1, the upper limit of mains power output in the distribution network system of Example 2 is set as the upper limit of power transmission of transformers T1 and T2 within the distribution network; such as Figure 6 and Figure 7As shown, the power output curve of the new energy power plant is based on the annual wind and solar power output data provided by the ENTSO platform. After normalization and scenario reduction, the typical power output curve of new energy used in this embodiment is obtained; as shown Figure 8 As shown, typical power load demand curves for each data center within the power distribution network system are defined.

[0162] In step S1, based on empirical values, the failure rate and repair rate of the components in the power distribution network system are set as shown in Table 1.

[0163] In step S1, the system average number of outages (SAIFI), system average outage time (SAIDI), expected load shedding (LOLE), and expected power shortage (EENS) are selected as evaluation indicators for the reliability assessment method.

[0164] In steps S2 to S3, the key parameters used in the reliability assessment method of the power distribution network system are set as shown in Table 2.

[0165] In steps S4 to S5, the parameters set for the centralized backup power storage configuration optimization model and the parameters set for solving the model are shown in Table 3.

[0166] Table 3 Parameters and solution parameters of the centralized backup power storage configuration optimization model

[0167] parameter numerical values <![CDATA[Energy capacity cost σ1 (yuan / kWh)]]> 600 <![CDATA[Power capacity cost σ2 (yuan / kW)]]> 300 <![CDATA[Lower limit β1 of E / C]]> 3 <![CDATA[Upper limit β2 of E / C]]> 8 <![CDATA[Number of particles N p > 32 <![CDATA[Number of iterations N iter > 100

[0168] In step S5, the adaptive particle swarm optimization algorithm is selected to solve the proposed centralized backup power storage configuration optimization model.

[0169] In this embodiment, two scenarios are set up in the implementation plan:

[0170] ①Scenario 1 - Centralized energy storage is configured in a shared mode in the data center within the cluster;

[0171] ②Scenario 2 - Each data center is configured with independent energy storage.

[0172] Regarding reliability requirements, it is assumed that each data center in the data center cluster has the same reliability requirements, namely SAIFI≤0.5h / a, SAIDI≤0.4h / a, LOLE≤24h / a, and EENS≤24MWh / a.

[0173] By implementing the above scheme, the optimal configuration capacity of backup power storage can be obtained for the two scenarios. The optimized configuration results of backup power storage for scenario 1 and scenario 2 are shown in Appendix Tables 4 and 5.

[0174] Table 4 Optimization configuration results for Scenario 1 (Centralized Energy Storage)

[0175] <![CDATA[Energy capacity E SESS (kWh)]]> <![CDATA[Power capacity C SESS (kW)]]> Cost (ten thousand yuan) 13800 3540 934.20

[0176] Table 5 Optimization configuration results for Scenario 2 (independent energy storage)

[0177] main body <![CDATA[Energy capacity E SESS (kWh)]]> <![CDATA[Power capacity C SESS (kW)]]> Cost (ten thousand yuan) Data Center 1 18500 2320 1179.60 Data Center 2 27200 3420 1734.60 Data Center 3 37800 4740 2410.20 total 83500 10480 5324.40

[0178] The energy storage configuration results for the two scenarios were verified using the proposed reliability assessment module. It was found that the reliability indicators of each data center in the distribution network met the proposed minimum reliability requirements. This fully demonstrates that the designed energy storage capacity optimization configuration algorithm can search for suitable energy storage configuration schemes while meeting reliability requirements.

[0179] By comparing the energy storage configuration results of Scenario 1 and Scenario 2, it was found that under the same reliability requirements, the shared energy storage capacity is smaller than the independent energy storage configuration capacity. From the overall configuration perspective, the required shared energy storage capacity is 83.5% lower than the total capacity of independent energy storage, the power capacity is 66.2% lower, and the configuration cost is 82.3% lower. This shows that under the same reliability requirements, the cost of shared centralized energy storage is much lower than that of independent energy storage solutions.

[0180] Meanwhile, the results also revealed that when choosing a shared centralized energy storage configuration, the energy capacity requirement for energy storage is relatively small. This is because when a data center fails, other data centers can still supply power to that data center through the redundant capacity of their power supply and distribution systems, thereby reducing the demand for energy storage power station capacity to a certain extent.

[0181] Furthermore, by visually representing all the energy storage configurations searched during the solution process in Scenario 1, along with their corresponding comprehensive configuration costs, we can gain a better understanding of the solution process.

[0182] Following the above scheme, a scatter plot of the comprehensive energy storage configuration during the search process can be obtained as follows: Figure 10 As shown, the fitted surface of the overall energy storage configuration cost can be obtained as follows: Figure 11 As shown. In Figure 10 In the diagram, the position of each point represents the energy storage configuration searched by the particle swarm optimization algorithm during the search process, and the color of the point represents the overall configuration cost of the corresponding energy storage configuration; Figure 11 In the diagram, the distribution of the surface along the Z-axis and the color of the surface represent the level of configuration cost.

[0183] Depend on Figure 10It is evident that, on the one hand, the distribution of the scattered points clearly shows that the energy storage configuration points are all within the area enclosed by several straight lines, reflecting the "hard" feasible region formed by the constraint relationship between energy storage capacity and power. On the other hand, judging from the color of the scattered points, there is a dark orange / dark red area in the lower left corner of the image. This area represents energy storage configuration points with very low configuration costs, but due to not meeting reliability requirements, the penalty coefficient is large, resulting in a high overall configuration cost and forming a reliability-infeasible region. At the same time, judging from the clustering of the scattered points, a large number of energy storage configuration points are clustered in the lower left corner of the reliability feasible region, that is, in the area that meets reliability requirements and has relatively low configuration costs. This indicates that the algorithm has searched this area thoroughly, and the optimal result also converges to the lower left corner of this area.

[0184] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be within the scope of protection of the present invention.

Claims

1. A method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture, characterized in that, Includes the following steps: Step S1: Collect historical or planned operational data of the target data center cluster system; Step S2: Construct a fault management module; Step S3: Construct a reliability assessment module based on the fault management module; the reliability assessment module is used to perform fault timing simulation. When a fault occurs, the fault management module is used to determine the power supply path and power supply of each data center, and finally obtain the reliability value of each data center. The reliability is a combination of one or more power distribution network reliability indicators; Step S4: Construct a centralized backup power storage configuration optimization model based on the reliability assessment module; the centralized backup power storage configuration optimization model uses the energy capacity and power capacity of the centralized backup power storage power station as energy storage configuration parameters, reliability constraints and energy storage configuration constraints as constraints, and minimizing energy storage cost as the objective function; the combination function of the reliability penalty function and the objective function is used as the fitness function. The reliability constraint is expressed as follows: r i (E SESS ,C SESS )≤r i req r∈R In the formula, R is the set of reliability indices, r∈R is a specific reliability index, and r i (E SESS C SESS ) represents the reliability value under the corresponding energy storage configuration, r i req The preset reliability requirements for data center i; The reliability penalty function P(r) is expressed by the following formula: In the formula, r∈R is a specific reliability index, r req λ is the preset reliability requirement value for data center i, and λ is the penalty coefficient. Step S5: Based on the historical operating data, solve the centralized backup power storage configuration optimization model using the fitness function to obtain the optimal energy storage configuration parameters.

2. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 1, characterized in that: The target data center cluster system includes a centralized backup power storage station, at least one power source, and at least one data center. The target data center cluster system adopts a cluster-shared architecture: the centralized backup power storage station includes energy storage batteries and energy storage buses; each data center includes a power supply and distribution system, a computer room, and a DC bus; the power supply and distribution system of each data center uses DC to supply power to the computer room, and the DC bus of each data center is connected to the energy storage bus of the centralized backup power storage station through DC cables, and the energy storage bus of the centralized backup power storage station is connected to the energy storage batteries through DC cables.

3. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 1, characterized in that: The fault management module includes a power supply path confirmation unit and a multi-entity power allocation unit. The power supply path confirmation unit is used to obtain the power supply path of the data center and is configured to determine the power supply path of each data center by first using power supply, then using centralized energy storage, and finally using centralized energy storage backup energy. The multi-entity power distribution unit is used to obtain the power supply of the data center and is configured as follows: for the target data center cluster system or electrical island, if the total maximum power supply capacity is greater than or equal to the total power demand of the data center, it is determined that the power output is sufficient and the power supply of each data center is equal to the power demand. If the total maximum power supply capacity is less than the total power demand of the data center, it is determined that the power output is insufficient. The total maximum power supply capacity is then divided according to the ratio of the power demand of each data center to obtain the power supply capacity of each data center.

4. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 1, characterized in that: The energy storage configuration constraints include upper and lower limits for energy capacity, upper and lower limits for power capacity, and upper and lower limits for the ratio of energy capacity to power capacity.

5. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 1, characterized in that: In step S5, a heuristic algorithm is used for iterative solution. During the iterative solution process, for each set of energy storage configuration parameters, the reliability assessment module is used to perform fault timing simulation to obtain the reliability value of each data center. The reliability value of each data center and the energy storage cost corresponding to the energy storage configuration parameters are input into the fitness function to obtain the fitness value.

6. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 5, characterized in that: The reliability assessment module is used to perform M failure timing simulations, and the average value of the M original reliability values ​​for each data center is taken to obtain the reliability value. Each fault timing simulation includes: based on the failure rate and repair rate of each component, using the sequential Monte Carlo method, randomly generating a composite fault timing sequence of the system; then, according to the composite fault timing sequence of the system, under the centralized backup power storage configuration corresponding to the energy storage configuration parameters, combined with the typical power load curves of each data center and the typical output curves of each power source, performing fault timing simulation; and obtaining the original reliability value of each data center based on the timing simulation data.

7. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 5, characterized in that: In each simulation time step of the fault timing simulation, if a component is faulty or the backup power capacity of the centralized backup power storage station is less than the preset backup power capacity threshold, the fault management module is used to obtain the power supply path and power of each data center in the target data center cluster system through a two-stage calculation method, and runs to the next simulation time step according to the power supply path and power; otherwise, the system maintains the normal power supply path and power and runs to the next simulation time step.

8. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 7, characterized in that: The two-stage calculation method is as follows: Phase 1: The minimum power load of each data center is taken as the power demand; the power supply path confirmation unit obtains the temporary power supply path and power supply of each data center based on the power demand of each data center and the power supply capacity of each power supply; it is determined whether the power supply of each data center can meet the power demand: if the power demand of the data center is met, the normal power load of the data center is taken as the new power demand; otherwise, the data center is considered to be in a power outage. Phase Two: Using the power supply path confirmation unit, the power supply path and power output of each data center are obtained based on the new power demand of each data center and the power supply capacity of each power source.

9. The method for optimizing the configuration of centralized backup power storage in data centers based on a cluster-shared architecture as described in claim 8, characterized in that: The power supply path confirmation unit obtains the power supply path and power of each data center through the following process: First, the power supply is used as the power source to supply power to each data center according to the first power supply path. After using the connectivity algorithm to determine whether there is an electrical island in the target data center cluster system without considering centralized backup power storage facilities, for the target data center cluster system or each electrical island, the multi-main power distribution unit is used to calculate the first power supply power of each data center according to the power demand of each data center, and obtain the remaining power demand of each data center. Secondly, the power supply and centralized backup energy storage station are used as power supply sources to supply power to the data centers whose power demand is not met according to the second power supply path and the third power supply path, respectively. Using the multi-entity power distribution unit, the second power supply power of each data center whose power demand is not met is calculated according to the remaining power demand of each data center, and the load shedding of each data center is calculated. Finally, the power source is used as the power supply to charge the centralized backup energy storage station according to the fourth power supply path, thereby obtaining the charging power of the energy storage battery.

10. The method for optimizing the configuration of centralized backup power storage in a data center based on a cluster-shared architecture as described in claim 9, characterized in that: The first power supply path is: power supply - data center power distribution system - computer room; the second power supply path is: power supply - redundant power distribution system - energy storage bus - computer room; the third power supply path is: energy storage battery - energy storage bus - computer room; the fourth power supply path is: power supply - redundant power distribution system - energy storage bus - energy storage battery.

Images