Data center site selection planning method and system considering decision dependence on uncertainty

Abstract

The application discloses a data center site selection planning method and system considering decision-dependent uncertainty, and relates to the technical field of economic operation and planning of power systems. The application comprises: constructing a data center collaborative planning framework considering space-time transfer characteristics; establishing a decision-dependent uncertainty set for data center site selection planning; based on the decision-dependent uncertainty set, constructing a two-stage robust optimization model considering decision-dependent uncertainty; using an improved Benders&CCG algorithm to solve the two-stage robust optimization model to generate an optimal construction scheme for data center site selection. In the process of data center site selection planning, the application fully considers endogenous and exogenous uncertainty in the process, quantifies the endogenous correlation between decision and uncertainty in the construction process, avoids conservative solutions caused by traditional exogenous uncertainty sets, maintains the feasibility of operation adjustment in highly uncertain scenarios, and realizes global optimization of data center site selection planning.

Description

Technical Field

[0001] This invention relates to the field of power system economic operation and planning technology, specifically to a data center site selection planning method and system that takes into account the uncertainty of decision-making. Background Technology

[0002] As the physical carrier of digital infrastructure, modern data centers are experiencing exponential growth in computing power density and equipment deployment scale. However, this explosive growth has directly led to a surge in energy demand: according to statistics from Lawrence Berkeley National Laboratory, in 2020, US data centers consumed 140 billion kWh of electricity, accounting for 1.8% of the country's total electricity consumption. Data centers are crucial for supporting the normal operation of important sectors such as healthcare and transportation, and are the infrastructure for governments and enterprises to achieve digital transformation and intelligent services. The ability to rationally plan the site selection for data center construction has become a core contradiction that urgently needs to be addressed, hindering the sustainable development of related industries. Against this backdrop, scientifically planning data center site selection can not only enhance the supporting capabilities of information technology but also contribute to the high-quality development of the digital economy and intelligent society.

[0003] Currently, in fields like data center site selection and planning, which involve high capital sinks and extremely low fault tolerance, robust optimization, with its "defensive decision-making" characteristics, is more applicable to engineering. However, with the explosive growth of data processing volume, traditional robust methods based on static exogenous assumptions are gradually proving insufficient in the face of highly uncertain factors such as energy price fluctuations and policy adjustments. Research shows that in current practical engineering problems, decision-making often has a certain impact on uncertainty, known as decision-dependent uncertainty (DDU), or endogenous uncertainty caused by decisions. DDU constructs a dynamic relationship between decisions and the uncertainty set. For such engineering problems containing DDU, traditional robust methods with an invariant uncertainty set are no longer applicable and may lead to serious long-term economic losses due to decision-making errors. Data center planning requires determining the location, capacity, configuration, and connectivity of the data center with nodes requiring computing power. The location and connectivity decisions often affect the uncertainty of the computing power processing volume during data center operation. During operation, operational strategies need to be dynamically adjusted based on uncertain factors such as actual computing power demand. Compared to traditional decision-independent uncertainty (DIU) problems, DDU can more accurately characterize the risks of endogenous uncertainty, avoid the chain reaction of risks, and ensure that the response measures during the data center operation phase are always effective. Summary of the Invention

[0004] The purpose of this invention is to provide a data center site selection planning method and system that takes into account decision-dependent uncertainty. In the process of data center site selection planning, the endogenous and exogenous uncertainties in the site selection planning process are fully considered to ensure that the data center can dynamically respond to emergencies during operation. In the construction process, the endogenous correlation between decision and uncertainty needs to be fully quantified to avoid conservative solutions caused by traditional exogenous uncertainty sets, so as to maintain the feasibility of operational adjustments in highly uncertain scenarios, thereby achieving global optimization of computing power, energy and geographical resources in the process of data center site selection planning.

[0005] According to a first aspect of the present invention, in order to achieve the above-mentioned objective, the present invention provides the following technical solution: a data center site selection planning method considering decision-dependent uncertainty, comprising the following steps:

[0006] Construct a collaborative planning framework for data centers that takes into account spatiotemporal transfer characteristics;

[0007] Using the data center collaborative planning framework as the physical carrier, a decision dependency uncertainty set for data center site selection planning is established;

[0008] Based on the aforementioned decision dependency uncertainty set, a two-stage robust optimization model considering decision dependency uncertainty is constructed.

[0009] An improved Benders & CCG algorithm is used to solve the two-stage robust optimization model and generate the optimal construction plan for data center site selection.

[0010] Furthermore, the data center collaborative planning framework, which considers the spatiotemporal transfer characteristics, specifically includes:

[0011] Construction phase planning: Construction phase costs include land costs C L Fiber optic equipment construction cost C f ;

[0012] Operational Phase Planning: Operational phase costs include electricity costs C e , Load transfer bandwidth and traffic cost C bf ;

[0013] From a system architecture perspective, the data center collaborative planning framework includes two types of heterogeneous geographical node system architectures: the data center computing power demand node set and the data center construction node set.

[0014] Furthermore, using the data center collaborative planning framework as the physical carrier, a decision dependency uncertainty set for data center site selection planning is established, as follows:

[0015] (31) There are decision-related uncertainties in the site selection and construction of data centers. The set of decision-related uncertainties is expressed by the following formula:

[0016]

[0017] in, d i 0 represents the lower limit of the data processing volume of node i when DDU is not considered; This indicates the upper limit of data processing volume for node i when DDU is not considered; and d j These represent the upper and lower limits of uncertainty introduced when connecting node j, which requires computing power in the data center;

[0018] D Load As expressed by the following formula:

[0019]

[0020] In the formula, D Load i represents the lower limit of the total data processing volume connected to node i. This indicates the maximum total amount of data that node i can process.

[0021] (32) The carbon emissions of data centers located in different locations are also uncertain, which in turn affects the carbon emission costs of data centers, as expressed by the following formula:

[0022] EF i =EF fossil ×(1-p i,renewable )+EF renewable ×p i,renewable (3)

[0023] V i E =EF i ·(m i p se +n i p st )τc ele,i T (4)

[0024] In the formula, EF i p represents the carbon emission factor of node i; i,renewable Indicates the proportion of renewable energy in node i; EF fossil EF represents the carbon emission factor of fossil fuels. renewable The carbon emission factor represents renewable energy; VEi is the carbon emission of data center i; m i n i These represent the number of servers and storage devices at node i, respectively; p se p represents the operating power of a single server. st The power consumption of a single memory unit is τ, where τ is the PUE value of the data center, and c is the operating power of the memory unit. ele,iThe electricity price for building node i in the data center, where T is one year.

[0025] Furthermore, based on the aforementioned decision dependency uncertainty set, a two-stage robust optimization model considering decision dependency uncertainty is constructed, as follows:

[0026] The initial construction and subsequent operation of a data center is a two-stage issue in terms of time scale. The initial construction costs include land costs, construction engineering costs, equipment configuration costs, and communication fiber optic construction costs. The operation costs in the second stage after construction are completed include electricity costs, operation and maintenance costs, carbon emission costs, labor costs, bandwidth and traffic costs for load transfer, and penalty costs incurred when computing power demand cannot be met.

[0027] The objective function is to minimize the site selection planning cost of data centers considering load shifting capabilities. In order to more accurately simulate the various costs of data center construction, the planning cost is calculated as the total cost of building a data center for one year. The construction cost is amortized using the equal annual value method. The carbon emission factor is divided into multiple segments within one year for fitting, and it is assumed that each computing power demand node is connected to only one data center.

[0028] The objective function includes the equivalent annual construction cost and annual operating cost of the data center. The objective function for establishing the two-stage robust model for data center construction and investment is shown below:

[0029]

[0030] In the formula: C bu C is the equivalent annual construction cost of the data center; op The annual operating cost of the data center; the first-stage decision variable includes a i With b ij , where a i b is a 0-1 variable representing whether the data center at location i is planned to be built. ij d is a 0-1 variable indicating whether there is a connection between data center i and computing power demand node j; i,ini The sum of data received from connected computing power demand nodes before load transfer at node i data center represents the main manifestation of decision-related uncertainty; d i For the second-stage decision variable, it represents the amount of data that needs to be processed after the data center at node i transfers its load.

[0031] Furthermore, the equivalent annual construction cost C bu Specifically, it includes:

[0032] (51) Land Costs

[0033] The land acquisition cost for data center construction nodes is determined by the land transfer price per unit area in the node region and the land area occupied by the data center, as shown in the following formula:

[0034]

[0035] In the formula: C L Total land cost for the data center; L p,i L is the unit price for land transfer. s,i ω0 represents the land area, ω0 represents the benchmark discount rate, y represents the investment amortization period, I represents the set of nodes planned for the data center construction, and a represents the area of ​​land occupied. i A 0-1 variable representing whether the data center at planned construction site i is to be built;

[0036] (52) Construction costs

[0037] The construction cost of a data center includes the configuration costs and building installation costs of various systems within the data center, including power supply and distribution systems, cooling systems, and internal communication systems, as shown in the following formula:

[0038]

[0039] In the formula: C co For the total construction cost of the data center; F s,i The cost of constructing the data center for node i;

[0040] (53) Equipment configuration costs

[0041] Equipment configuration costs refer to the cost of configuring servers and storage in a data center. The number of servers and storage units is determined by the amount of data the data center needs to process, as shown in the following formula:

[0042]

[0043] In the formula: C eq Total equipment configuration cost for the data center; m i n i C represents the number of servers and storage devices for node i, respectively. se With C st These are the unit prices for servers and storage, respectively.

[0044] (54) Cost of fiber optic cable construction

[0045] The cost of constructing fiber optic cables for data centers consists of two parts: the cost of constructing fiber optic cables connecting data centers to each other, and the cost of constructing fiber optic cables connecting computing power-demanding nodes to the data center, as shown in the following formula:

[0046]

[0047] In the formula: C f C is the total cost of constructing fiber optic communication cables for the data center. opf D is the unit price for constructing optical fiber for communication. ik D is the distance between two data center nodes. ij The distance between the data center and the nodes requiring computing power;

[0048] In summary, the formula for the equivalent annual cost of data center construction is as follows:

[0049] C bu =C L +C co +C eq +C f (10).

[0050] Furthermore, the annual operating cost C of the data center op Specifically, it includes:

[0051] (61) Electricity Costs

[0052] The servers and storage configured in the data center need to meet the received data center requirements, so they must be kept fully operational. In addition, servers and other IT equipment generate heat, necessitating the operation of cooling systems. The power consumption of the cooling system and other equipment is determined by the data center's power efficiency rating. Therefore, the data center's electricity costs are as follows:

[0053]

[0054] In the formula: C e p represents the total electricity cost of the data center. se p represents the operating power of a single server. st The power consumption of a single memory unit is τ, where τ is the PUE value of the data center, and c is the operating power of the memory unit. ele,i The electricity price for building node i in the data center, where T is one year;

[0055] (62) Operation and maintenance costs

[0056] Data center operation and maintenance costs refer to the expenses incurred in maintaining and servicing various equipment during the operation of the data center. This cost is calculated by multiplying the data center's equipment configuration and construction costs by a fixed percentage, as detailed below:

[0057] C ma =η(C co +C eq (12)

[0058] In the formula: C ma η represents the total operation and maintenance cost of the data center, and η is the proportion of operation and maintenance cost.

[0059] (63) Carbon emission costs

[0060] The carbon emission costs of data centers are as follows:

[0061]

[0062] In the formula: C co2 It is the total carbon cost of data centers, c co2,i V is the carbon emission unit price of data center construction node i. i E λ represents the carbon emissions of data center i, α represents the base carbon price, l represents the length of the billing interval based on the base carbon price, and as the carbon emissions of data center i increase, the unit carbon price gradually increases.

[0063] (64) Labor costs

[0064] The labor costs of a data center are determined by the salary levels and number of employees. Employee salaries include basic salary, social security contributions, and housing provident fund contributions. The salary levels of personnel at different nodes vary and are specifically assessed based on the local GDP level, as shown in the following formula:

[0065]

[0066] In the formula: C w This is the total labor cost of the data center, q i The number of employees in data center i is w. i It represents the annual salary level of personnel at node i;

[0067] (65) Bandwidth and flow costs for load transfer

[0068] When a data center transfers received data processing requests to other node data centers via fiber optic cables, bandwidth costs are incurred. The bandwidth costs incurred by nodes requesting computing power when transmitting data to the data center are not borne by the data center but by the user. The unit price of bandwidth for data transmission varies between different nodes, as shown in the following formula:

[0069]

[0070] In the formula: C bf c is the total bandwidth cost for the data center. bf,i The unit price of bandwidth traffic for data center node i, d i,ini Let d be the sum of the data received by node i from the connected computing power demand nodes before the load is transferred. i This represents the amount of data that needs to be processed after the data center at node i transfers its load.

[0071] (66) Penalty Fees

[0072] If a data center cannot meet its computing power requirements during operation, it will incur certain penalty costs, as shown in the following formula:

[0073]

[0074] In the formula: C pu The penalty fee for the data center's inability to meet demand, d j Indicates the amount of computing power required by the computing power demand node; c pu The penalty cost per unit of computing power lost;

[0075] In summary, the formula for the annual operating cost of a data center is as follows:

[0076] C op =C e +C ma +C co2 +C w +C bf +C pu (18).

[0077] Furthermore, the constraints for the first and second phases are as follows:

[0078] (71) Constraints for the first stage:

[0079] (71.1) 0-1 variable constraints

[0080] Whether node i in the data center construction project will build a data center is variable a. i The variable b indicates whether there is a connection between data center i and computing power demand node j. ij All are 0-1 variables:

[0081] a i ∈{0,1} b ij ∈{0,1} i∈I,j∈J (19)

[0082] (71.2) Node connectivity constraints for computing power requirements

[0083] Each node requiring computing power can connect to one and only one data center, and must be connected to a node that has a data center in operation.

[0084]

[0085] (72) Constraints in the second stage

[0086] For a given d∈D(a i b ij The second phase has the following constraints:

[0087] (72.1) Data centers accept load constraints

[0088] Before the data center relocation, its load should be equal to the sum of the data volume of the computing power demand nodes connected to it. After the relocation, the sum of the loads of all data centers remains constant.

[0089]

[0090] u se m min ≤d i ≤u se m max,i (twenty four)

[0091] Where: m min,i This represents the minimum number of servers required for a data center; m max,i The maximum number of data center servers for node i;

[0092] (72.2) Data Center IT Equipment Constraints

[0093] The servers and storage devices in a data center need to be able to handle the amount of data they receive. Servers must be able to process all data arriving instantaneously, and storage devices must have sufficient storage capacity for the data retention period.

[0094]

[0095] In the formula: u se The data processing rate of a single server; u st Δt represents the data storage capacity of a single memory unit; Δt represents the time the data needs to be stored.

[0096] Furthermore, the improved Benders & CCG algorithm is used to solve the two-stage robust optimization model to generate the optimal data center site selection and construction plan, as follows:

[0097] (81) Since the maximum value and the multiplication of bilinear terms in formula (16) introduce a nonlinear model, the McCormick envelope is used to linearize the model. The nonlinearity handling strategy is as follows:

[0098] Since the total bandwidth cost includes a maximum value, a new variable z is introduced first. i Replacing the max term, equation (16) is transformed into the following equation:

[0099]

[0100] z i ≥d i,ini -d i (28)

[0101] z i ≥0 (29)

[0102] The McCormick envelope can relax a non-convex problem into a convex problem; therefore, using the McCormick envelope can reduce the bandwidth cost C of load transfer. bf The bilinear terms are transformed into a linear problem, and the final equation (16) is replaced by the following formula:

[0103]

[0104] (82) For the two-stage robust problem that takes into account the uncertainty of decision dependence, the improved Benders & CCG algorithm is used for solution. The improved Benders & CCG algorithm fully considers the uncertainty of decision dependence and introduces three sub-problems into the traditional CCG algorithm. Sub-problem 1 is used to determine whether the original problem has a solution. Sub-problems 2 and 3 are solved iteratively for the two cases of having a solution and not having a solution, respectively. The specific solution steps are as follows:

[0105] Step 1: Initialization

[0106] Let LB = -∞, UB = +∞, and define the set of poles π. Collection with polar rays γ It is an empty set;

[0107] Step 2: Solve the main problem

[0108] Solve the following main problem:

[0109]

[0110] In the formula, ω represents the objective function of the two-stage robust problem, c1 represents the coefficient matrix of the decision variables in the first stage, x represents the vector of the decision variables in the first stage, η represents the objective function value in the second stage, X represents the set of decision variables in the first stage, π represents the vector composed of poles, d-B1x-Eu represents the constraint of decision dependency uncertainty corresponding to the poles, OU represents the KKT conditions for x and π, λ represents the Lagrange multipliers corresponding to the poles in the KKT conditions, and P Π Let denot be the set of poles, γ be the vector composed of polar rays, d-B1x-Ev be the constraint on the decision-dependent uncertainty corresponding to the polar ray, OV be the KKT condition for v and γ, ζ be the Lagrange multiplier corresponding to the polar ray in the KKT condition, and R be the set of poles. Π Represents a collection of polar rays;

[0111] Solving for x* and η yields solutions, where η is a two-stage max-min problem. The lower bound LB is then updated to ω.

[0112] Step 3: Solving Subproblem 1

[0113] Solving subproblem 1 yields the uncertainties u*f and η. f (x*):

[0114]

[0115] In the formula, η f (x*) represents the objective function value of subproblem 1, u represents the uncertainty variable, U(x*) represents the uncertainty set that the decision depends on, and y represents the decision variable of the second stage;

[0116] Step 4: Regarding η f (x*) Classification of results

[0117] If η f (x*)=0

[0118] Calculate subproblem 2 to obtain η s (x*), u*s, and the corresponding pole π*:

[0119]

[0120] In the formula, η s (x*) represents the objective function value of subproblem 2, c2y represents the objective function of the second stage problem, y is the vector set of decision variables in the second stage, and Y(x*,u) represents the constraints in the second stage.

[0121] renew Substitute the following two constraints into the main problem:

[0122] η≥(π*) T d-(π*) T B1x-(π*) T Eu π* (35)

[0123] (u π* ,λ π )∈OU(x,π*)(36)

[0124] If η f (x*)>0

[0125] Calculate subproblem 3, obtain the polar ray γ*, and set η. s (x*)=+∞:

[0126]

[0127] renew Substitute the following two constraints into the main problem:

[0128] (γ*)T d-(γ*) T B1x-(γ*) T Ev γ* ≤0 (38)

[0129] (v γ* ,ζ γ* )∈OV(x,γ*)(39)

[0130] Step 5: Update UB = min{UB, c1x* + η} s (x*)}

[0131] Step 6: If UB-LB≤δ, obtain x* and terminate the operation; otherwise, t=t+1 and return to step 2 to continue the iteration.

[0132] According to a second aspect of the present invention, the present invention provides a data center site selection planning system that takes into account decision-dependent uncertainty, for implementing the data center site selection planning method that takes into account decision-dependent uncertainty described in the first aspect, comprising:

[0133] The building module is used to construct a collaborative planning framework for data centers that takes into account spatiotemporal transfer characteristics;

[0134] The decision dependency uncertainty set establishment module is used to establish the decision dependency uncertainty set for data center site selection planning, using the data center collaborative planning framework as the physical carrier.

[0135] The model building module is used to construct a two-stage robust optimization model that considers the uncertainty of decision dependencies based on the decision dependency uncertainty set.

[0136] The model solution output module uses the improved Benders & CCG algorithm to solve the two-stage robust optimization model and generate the optimal construction plan for data center site selection.

[0137] According to a second aspect of the present invention, a terminal device is provided, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor loads and executes the computer program, it employs the data center site selection planning method that takes into account decision-dependent uncertainty as described in the first aspect.

[0138] This invention has at least the following beneficial effects:

[0139] (1) By constructing a decision-dependent uncertainty set, this invention accurately quantifies the endogenous relationship between site selection and computing power requirements, thus avoiding the conservative solution caused by the pre-set fixed uncertainty set in traditional robust optimization.

[0140] (2) The method proposed in this invention achieves global optimization of computing power demand and energy resources through load transfer mechanism and collaborative planning framework that considers DDU. In extreme scenarios, data centers are preferentially located in nodes with low electricity prices and low carbon emissions, and load allocation is flexibly adjusted through bandwidth transfer, which not only reduces operating costs, but also responds to the strategic needs of green computing in the "East Data West Computing" project.

[0141] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0142] Figure 1 This is a flowchart illustrating the planning method described in this invention;

[0143] Figure 2 This is a schematic diagram of the data center collaborative planning framework of the present invention;

[0144] Figure 3 This is a schematic diagram illustrating the cost composition of the data center construction project of this invention;

[0145] Figure 4 This is a schematic diagram illustrating the cost composition of data center planning, considering whether or not DDU is taken into account in this invention. Detailed Implementation

[0146] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0147] Please see Figures 1-4 This invention provides a technical solution: a data center site selection and planning method considering decision-dependent uncertainty, comprising the following steps:

[0148] (1) A data center collaborative planning framework considering spatiotemporal transfer characteristics is detailed below:

[0149] Data center lifecycle planning needs to take into account the coupled impact of the construction and operation phases. Construction phase costs include land costs (C). L Fiber optic equipment construction cost C f Construction expenditures, and operating costs including electricity costs C e , Load transfer bandwidth and traffic cost C bfOperating expenses, etc. It's worth noting that since data center processing typically involves critical loads, the data centers selected and constructed must still be able to meet computing power demands even in the worst-case scenario. The planning scheme must meet reliability constraints under extreme scenarios, meaning that when facing the most severe demand fluctuations, the deployed data centers can still maintain a certain level of data processing capacity and exhibit strong robustness. Furthermore, site selection planning decisions must consider both integer decision variables related to data center construction and connectivity, and continuous decision variables such as data center computing power configuration. In essence, the site selection planning problem is a mixed-integer two-stage robust optimization model.

[0150] From a system architecture perspective, data center site selection planning networks comprise two types of heterogeneous geographical nodes: the first is the set of data center demand nodes (Demand Node, DN) J, where the data demands processed by the data center do not all originate locally but are transmitted from other regions with data processing needs. The second is the set of data center construction nodes (Data Center Node, DCN) I, representing the planned construction locations of the data centers. Different locations have different unit costs for land, electricity, and personnel, and are subject to certain resource constraints. Fiber optic network topology not only requires establishing service access links from demand nodes to data centers but also constructing collaborative data processing channels between data centers. The interconnection between data centers fundamentally alters the locally optimal nature of traditional single-center planning. The designed collaborative planning framework for data centers is as follows: Figure 2 As shown.

[0151] Further analysis of the disturbance mechanism of data center computing power fluctuations on system robustness reveals that due to the inherent fluctuations in computing power demand nodes, there is an endogenous correlation between the data center's computing power and planning decision variables. That is, different data center construction and fiber optic connection decisions affect the uncertainty of data center computing power. This means that traditional robust optimization methods that pre-set a fixed uncertainty set are no longer applicable. A two-stage min-max-min optimization framework with recursive feedback characteristics is required. The first stage determines the site selection scheme to minimize the total cost; the second stage dynamically adjusts the uncertainty set based on the decision variables and then optimizes the real-time load scheduling strategy based on the uncertainty set to cope with the worst-case scenario. Therefore, data center site selection and construction is essentially a two-stage robustness problem considering DDU (Data Usage Daily Demand).

[0152] (2) The decision-making process for data center site selection planning depends on the construction of an uncertain set, which is explained in detail below:

[0153] Data center site selection and planning are subject to a series of uncertainties. To ensure that the planning results meet the computing and storage needs of different scenarios, these uncertainties are categorized into endogenous and exogenous uncertainties. Endogenous uncertainties are primarily determined by the data center's site selection and construction, as well as its connectivity with nodes requiring computing power. Specifically, due to the uncertainty of computing power demand, the location of the data center and its connectivity with these nodes directly affect the uncertainty of the data processed by the data center; that is, data center site selection and construction involve decision-related uncertainties. The set of decision-related uncertainties can be expressed by the following formula:

[0154]

[0155] in, d i 0 represents the lower limit of the data processing volume of node i when DDU is not considered; This indicates the upper limit of data processing volume for node i when DDU is not considered; and d j Let D represent the upper and lower limits of uncertainty introduced by node j when the data center connects to the computing power demand node, respectively, and D Load It can be expressed by the following formula:

[0156]

[0157] In the formula, D Load i represents the lower limit of the total data processing volume connected to node i. This represents the total upper limit of the data processing volume connected to node i.

[0158] Exogenous uncertainty is primarily determined by external factors. Due to variations in renewable energy output across different locations and under varying weather conditions, the carbon emissions of data centers located in different locations also exhibit uncertainty, thus affecting their carbon emission costs. This can be specifically expressed by the following formula:

[0159] EF i =EF fossil ×(1-p i,renewable )+EF renewable ×p i,renewable (42)

[0160] V i E =EF i ·(m i p se +n i p st )τc ele,i T (43)

[0161] In the formula, EF i p represents the carbon emission factor of node i; i,renewable Indicates the proportion of renewable energy in node i; EF fossil EF represents the carbon emission factor of fossil fuels. renewable The carbon emission factor represents renewable energy; VEi is the carbon emission of data center i; m i n i These represent the number of servers and storage devices at node i, respectively; p se p represents the operating power of a single server. st The power consumption of a single memory unit is τ, where τ is the PUE value of the data center, and c is the operating power of the memory unit. ele,i The electricity price for building node i in the data center, where T is one year;

[0162] (3) Data center site selection planning methods that consider decision-dependent uncertainty are detailed below:

[0163] The initial construction and subsequent operation of a data center are two phases in terms of time. The initial construction costs consist of land costs, construction engineering costs, equipment configuration costs, and communication fiber optic cable construction costs. The second phase of operation costs after construction is completed consists of electricity costs, operation and maintenance costs, carbon emission costs, labor costs, bandwidth and traffic costs for load shifting, and penalty costs incurred when computing power demands cannot be met.

[0164] The objective function is to minimize the site selection planning cost of the data center considering load shifting capabilities. In order to more accurately simulate the various costs of data center construction, the planning cost is calculated as the total cost of building the data center for one year. The construction cost is amortized using the equal annual value method. The carbon emission factor is divided into multiple segments within one year for fitting, and it is assumed that each computing power demand node is connected to only one data center.

[0165] The objective function includes the equivalent annual construction cost and annual operating cost of the data center. The objective function for establishing the two-stage robust model for data center construction and investment is shown below:

[0166]

[0167] In the formula: C bu C is the equivalent annual construction cost of the data center; op The annual operating cost of the data center; the first-stage decision variable includes a i With b ij , where a i b is a 0-1 variable representing whether the data center at location i is planned to be built. ij d is a 0-1 variable indicating whether there is a connection between data center i and computing power demand node j; i,iniThe sum of data received from connected computing power demand nodes before load transfer at node i data center represents the main manifestation of decision-related uncertainty; d i For the second-stage decision variable, it represents the amount of data that needs to be processed after the data center at node i transfers its load.

[0168] 1. The objective function for the first stage – equivalent annual construction cost

[0169] 1) Land costs

[0170] The land acquisition cost for data center construction nodes is determined by the land transfer price per unit area in the node region and the land area occupied by the data center, as shown in the following formula:

[0171]

[0172] In the formula: C L Total land cost for the data center; L p,i L is the unit price for land transfer. s,i ω0 represents the land area, ω0 represents the benchmark discount rate, y represents the investment amortization period, and I represents the set of nodes planned for the data center construction.

[0173] 2) Construction costs

[0174] The construction cost of a data center includes the configuration costs and building installation costs of various systems within the data center, including power supply and distribution systems, cooling systems, internal communication systems, etc., as shown in the following formula:

[0175]

[0176] In the formula: C co For the total construction cost of the data center; F s,i The cost of building the data center for node i.

[0177] 3) Equipment configuration costs

[0178] Equipment configuration costs refer to the cost of configuring servers and storage in a data center. The number of servers and storage units is determined by the amount of data the data center needs to process, as shown in the following formula:

[0179]

[0180] In the formula: C eq Total equipment configuration cost for the data center; m i n i C represents the number of servers and storage devices for node i, respectively. se With C st These are the unit prices for servers and storage, respectively.

[0181] 4) Costs of constructing optical fiber communication cables

[0182] The cost of constructing fiber optic cables for data centers consists of two parts: the cost of constructing fiber optic cables connecting data centers to each other, and the cost of constructing fiber optic cables connecting computing power-demanding nodes to the data center, as shown in the following formula:

[0183]

[0184] In the formula: C f C is the total cost of constructing fiber optic communication cables for the data center. opf D is the unit price for constructing optical fiber for communication. ik D is the distance between two data center nodes. ij This refers to the distance between the data center and the nodes that require computing power.

[0185] In summary, the formula for the equivalent annual cost of data center construction is as follows:

[0186] C bu =C L +C co +C eq +C f (49)2. Objective function of the second stage – operating cost

[0187] 1) Electricity Costs

[0188] When a data center receives data processing requests from nodes requiring computing power, it needs to activate servers for processing and also activate storage for storage. The servers and storage configured in the data center must meet the received data center demands, therefore they must remain fully operational. Furthermore, servers and other IT equipment generate heat, necessitating the operation of cooling systems. To simplify computation, the power consumption of cooling systems and other equipment is determined by the data center's Power Usage Effectiveness (PUE) value. The resulting electricity costs for the data center are as follows:

[0189]

[0190] In the formula: C e p represents the total electricity cost of the data center. se p represents the operating power of a single server. st The power consumption of a single memory unit is τ, where τ is the PUE value of the data center, and c is the operating power of the memory unit. ele,i The electricity price for building node i in the data center, where T is one year.

[0191] 2) Operation and maintenance costs

[0192] Data center operation and maintenance costs refer to the expenses incurred in maintaining and servicing various equipment during the operation of the data center. They are generally calculated by multiplying the data center's equipment configuration and construction costs by a fixed percentage.

[0193] C ma =η(C co +C eq (51)

[0194] In the formula: C ma η represents the total operation and maintenance cost of the data center, and η is the proportion of operation and maintenance cost.

[0195] 3) Carbon emission costs

[0196] To promote low-carbon operation of data centers, it is proposed to measure and charge for the carbon emissions from the electricity consumed by data centers, thereby guiding the construction of data centers and the transfer of loads in western regions, and thus solving the problem of renewable energy consumption in western regions. The carbon emission costs for data centers are as follows:

[0197]

[0198] In the formula: C co2 It is the total carbon cost of data centers, c co2,i λ is the carbon emission unit price of data center construction node i, VE i is the carbon emission of data center i, λ is the base carbon price; α is the carbon emission unit price growth rate, and l is the length of the billing interval based on the base carbon emission unit price. As the carbon emission of data center node i increases, the unit carbon emission price gradually increases.

[0199] 4) Labor costs

[0200] The labor costs of a data center are primarily determined by the salary levels and number of employees. Employee salaries should include basic salary, social security contributions, and housing provident fund contributions. Salary levels vary at different nodes and can be assessed based on the local GDP level, as shown in the following formula:

[0201]

[0202] In the formula: C w This is the total labor cost of the data center, q i The number of employees in data center i is w. i It represents the annual income level of personnel at node i.

[0203] 5) Bandwidth and traffic costs for load transfer

[0204] When a data center transfers received data processing requests to other node data centers via fiber optic cables, bandwidth costs are incurred. The bandwidth costs incurred by nodes transmitting data to the data center are not borne by the data center, but by the user. The unit price of bandwidth for data transmission varies between different nodes, as shown in the following formula:

[0205]

[0206] In the formula: C bf c is the total bandwidth cost for the data center. bf,i The unit price of bandwidth traffic for data center node i, d i,ini Let d be the sum of the data received by node i from the connected computing power demand nodes before the load is transferred. i This represents the amount of data that needs to be processed after the load is transferred from node i to the data center.

[0207] 6) Penalty fees

[0208] If a data center fails to meet its computing power requirements during operation, it will incur certain penalty costs, as shown in the following formula:

[0209]

[0210] In the formula: C pu The penalty fee for the data center's inability to meet demand, d j Indicates the amount of computing power required by the computing power demand node; c pu The penalty cost per unit of computing power lost.

[0211] The formula for the annual operating cost of a data center is as follows:

[0212] C op =C e +C ma +C co2 +C w +C bf +C pu (57)3. Constraints in the first stage

[0213] 1) 0-1 variable constraints

[0214] Whether node i in the data center construction project will build a data center is variable a. i The variable b indicates whether there is a connection between data center i and computing power demand node j. ij All are 0-1 variables:

[0215] a i ∈{0,1} b ij ∈{0,1} i∈I,j∈J (58)

[0216] 2) Node connectivity constraints for computing power requirements

[0217] Each node requiring computing power can connect to one and only one data center, and must be connected to a node that has a data center in operation.

[0218]

[0219] 4. Constraints in the second phase

[0220] For a given d∈D(a i b ij The second phase has the following constraints:

[0221] 1) Data centers accept load constraints

[0222] Before the data center relocation, its load should be equal to the sum of the data volume of the computing power demand nodes connected to it. After the relocation, the sum of the loads of all data centers remains constant.

[0223]

[0224] u se m min ≤d i ≤u se m max,i (63)

[0225] Where: m min,i This represents the minimum number of servers required for a data center; m max,i This represents the maximum number of data center servers for node i.

[0226] 2) Data center IT equipment constraints

[0227] The servers and storage devices in a data center need to be able to handle the amount of data they receive. Servers must be able to process all data arriving instantaneously, and storage devices must have sufficient storage capacity for the data retention period.

[0228]

[0229] In the formula: u se The data processing rate of a single server; u st Δt represents the data storage capacity of a single memory unit; Δt represents the time the data needs to be stored.

[0230] At this point, the model for data center site selection planning considering decision-dependent uncertainty has been established. However, due to the presence of nonlinear cases and uncertainties with decision-dependent uncertainty in the model, the conventional CCG method for solving two-stage robustness is no longer applicable, and the model cannot be solved normally by conventional commercial solvers before simplification. Therefore, simplification methods are needed, and entirely new algorithms need to be designed to solve the two-stage robustness problem with decision-dependent uncertainty.

[0231] (4) The model solution scheme that takes into account the uncertainty of decision dependence is described in detail below:

[0232] The site selection and planning problem of data centers involves two stages: construction and operation. Decisions made during construction can significantly impact the uncertainty of computing power requirements during operation, making it a two-stage robust problem involving decision-related uncertainties. For such problems, the traditional CCG method is no longer applicable because the dynamic uncertainty set dependent on the decision may lead to unsolvable subproblems in the second stage. Therefore, this paper adopts an improved Benders & CCG algorithm to solve the two-stage robust problem considering decision-related uncertainties. Furthermore, since the maximum value and multiplication of bilinear terms in equation (16) introduce a nonlinear model, the McCormick envelope is used to linearize the model.

[0233] The nonlinearity handling strategy for the model is as follows:

[0234] Since the total bandwidth cost includes a maximum value, a new variable z is introduced first. i Replacing the max term, equation (16) is transformed into the following equation:

[0235]

[0236] z i ≥d i,ini -d i (67)

[0237] z i ≥0 (68)

[0238] The McCormick envelope can relax a non-convex problem into a convex problem; therefore, using the McCormick envelope can reduce the bandwidth cost C of load transfer. bf The bilinear terms are transformed into a linear problem, and the final equation (16) is replaced by the following expression:

[0239]

[0240] For two-stage robust problems that take into account the uncertainty of decision dependencies, the improved Benders & CCG algorithm can be used for solving them. The improved Benders & CCG algorithm fully considers the uncertainty of decision dependencies and introduces three subproblems into the traditional CCG algorithm. Subproblem 1 is used to determine whether the original problem has a solution. Subproblems 2 and 3 are solved iteratively for the cases of having a solution and not having a solution, respectively. The specific solution steps are as follows (written in simplified form):

[0241] Step 1: Initialization

[0242] Let LB = -∞, UB = +∞, and define the set of poles π. Collection with polar rays γ It is an empty set.

[0243] Step 2: Solve the main problem

[0244] Solve the following main problem:

[0245]

[0246] In the formula, ω represents the objective function of the two-stage robust problem, c1 represents the coefficient matrix of the first-stage decision variables, x represents the vector of the first-stage decision variables, η represents the objective function value of the second stage, X represents the set of the first-stage decision variables, π represents the vector composed of poles, d-B1x-Eu represents the constraint of decision dependency uncertainty corresponding to the poles, OU represents the KKT conditions for x and π, λ represents the Lagrange multipliers corresponding to the poles in the KKT conditions, and P Π Let denot be the set of poles, γ be the vector composed of polar rays, d-B1x-Ev be the constraint on the decision-dependent uncertainty corresponding to the polar ray, OV be the KKT condition for v and γ, ζ be the Lagrange multiplier corresponding to the polar ray in the KKT condition, and R be the set of poles. Π This represents a set of polar rays.

[0247] Solving for x* and η yields solutions, where η is a two-stage max-min problem. The lower bound LB is then updated to ω.

[0248] Step 3: Solving Subproblem 1

[0249] Solving subproblem 1 yields the uncertainties u*f and η. f (x*).

[0250]

[0251] In the formula, η f (x*) represents the objective function value of subproblem 1, u represents the uncertainty variable, U(x*) represents the uncertainty set that the decision depends on, and y represents the decision variable of the second stage.

[0252] Step 4: Regarding η f (x*) Classification of results

[0253] If η f (x*)=0

[0254] Calculate subproblem 2 to obtain η s (x*), u*s, and the corresponding pole π*.

[0255]

[0256] In the formula, η s (x*) represents the objective function value of subproblem 2, c2y represents the objective function of the second stage problem, y is the vector set of decision variables in the second stage, and Y(x*,u) represents the constraints of the second stage.

[0257] renew Substitute the following two constraints into the main problem:

[0258] η≥(π*) T d-(π*) T B1x-(π*) T Eu π* (74)

[0259] (u π* ,λ π )∈OU(x,π*)(75)

[0260] If η f (x*)>0

[0261] Calculate subproblem 3, obtain the polar ray γ*, and set η. s (x*) = +∞.

[0262]

[0263] renew Substitute the following two constraints into the main problem:

[0264] (γ*) T d-(γ*) T B1x-(γ*) T Ev γ* ≤0 (77)

[0265] (v γ* ,ζ γ* )∈OV(x,γ*)(78)

[0266] Step 5: Update UB = min{UB, c1x* + η} s (x*)}

[0267] Step 6: If UB-LB≤δ, obtain x* and terminate the operation; otherwise, t=t+1 and return to step 2 to continue the iteration.

[0268] At this point, the model solution scheme considering decision-dependent uncertainty has been established. The processed model can be effectively solved using a commercial solver, and the results are as follows: Figure 3 As shown in the figure. The results show that, compared with traditional robust methods, the proposed two-stage robust strategy that takes into account decision-dependent uncertainty can reduce the cost of data center site selection planning. The results are as follows. Figure 4 As shown.

[0269] In summary, this invention fully considers the endogenous and exogenous uncertainties in the data center site selection and planning process, ensuring that the data center can dynamically respond to unforeseen circumstances during operation. During construction, it is necessary to fully quantify the endogenous correlation between decision-making and uncertainty, avoiding conservative solutions caused by traditional exogenous uncertainty sets, in order to maintain the feasibility of operational adjustments in highly uncertain scenarios, thereby achieving global optimization of computing power, energy, and geographical resources in the data center site selection and planning process.

[0270] Example 2:

[0271] This embodiment provides a data center site selection planning system that takes into account decision-dependency uncertainty, used to implement the data center site selection planning method that takes into account decision-dependency uncertainty described in Embodiment 1, including:

[0272] The building module is used to construct a collaborative planning framework for data centers that takes into account spatiotemporal transfer characteristics;

[0273] The decision dependency uncertainty set establishment module is used to establish the decision dependency uncertainty set for data center site selection planning, using the data center collaborative planning framework as the physical carrier.

[0274] The model building module is used to construct a two-stage robust optimization model that considers the uncertainty of decision dependencies based on the decision dependency uncertainty set.

[0275] The model solution output module uses the improved Benders & CCG algorithm to solve the two-stage robust optimization model and generate the optimal construction plan for data center site selection.

[0276] Example 3:

[0277] This embodiment provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor loads and executes the computer program, it employs the data center site selection planning method that takes into account decision-dependent uncertainty as described in the first aspect.

[0278] It should be noted that the terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server, and the terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0279] Furthermore, the processor can be a central processing unit (CPU). Of course, depending on the actual use, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0280] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0281] For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible embodiments.

[0282] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0283] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A data center site selection and planning method that takes into account decision-dependent uncertainty, characterized in that, include; Construct a collaborative planning framework for data centers that takes into account spatiotemporal transfer characteristics; Using the data center collaborative planning framework as the physical carrier, a decision dependency uncertainty set for data center site selection planning is established; Based on the aforementioned decision dependency uncertainty set, a two-stage robust optimization model considering decision dependency uncertainty is constructed. An improved Benders & CCG algorithm is used to solve the two-stage robust optimization model to generate the optimal construction and deployment scheme for data center site selection. The constraints of the first and second stages of the two-stage robust optimization model are as follows: Constraints for Phase 1: 0-1 variable constraints Data center construction nodes i Variables of whether to build a data center Data Center i With computing power demand nodes j Whether the variables are connected All are 0-1 variables: （19） Computing power requirement node connection constraints Each node requiring computing power can connect to one and only one data center, and must be connected to a node that has a data center in operation. （20） （21） Second phase constraints For a given d ∈D( a i , b ij The second phase has the following constraints: Data centers accept load constraints Before the data center relocation, its load should be equal to the sum of the data volume of the computing power demand nodes connected to it. After the relocation, the sum of the loads of all data centers remains constant. （22） （23） （24） In the formula: m min,i This represents the minimum number of servers required for a data center. m max,i For nodes i Maximum number of data center servers; Data center IT equipment constraints The servers and storage devices in a data center need to be able to handle the amount of data they receive. Servers must be able to process all data arriving instantaneously, and storage devices must have sufficient storage capacity for the data retention period. （25） （26） In the formula: u se This refers to the data processing rate of a single server. u st This refers to the data storage capacity of a single memory unit; Δ t This refers to the time the data needs to be stored.

2. The data center site selection and planning method considering decision-dependent uncertainty according to claim 1, characterized in that: A collaborative planning framework for data centers that considers spatiotemporal transfer characteristics includes: Construction phase planning: Construction phase costs include land costs C L Construction project costs C co Equipment configuration cost C eq Fiber optic equipment construction cost C f ; Operational Phase Planning: Operational phase costs include electricity costs. C e Operation and maintenance costs C ma Carbon emission costs C co2 Labor costs C w Load transfer bandwidth and traffic costs C bf Penalty fee C pu ; From a system architecture perspective, the data center collaborative planning framework includes two types of heterogeneous geographical node system architectures: the data center computing power demand node set and the data center construction node set.

3. The data center site selection and planning method considering decision-dependent uncertainty according to claim 1, characterized in that: Using the data center collaborative planning framework as the physical carrier, a decision dependency uncertainty set for data center site selection planning is established, as follows: (31) There are decision-related uncertainties in the site selection and construction of data centers. The set of decision-related uncertainties is expressed by the following formula: （2） in, Indicates the node when DDU is not considered. i The lower limit of data processing volume; Indicates the node when DDU is not considered. i The upper limit of data processing volume; and These represent the nodes that require computing power to connect to the data center. j The uncertainty brought about by time has upper and lower limits; As expressed by the following formula: （3） In the formula, Represents a node i The lower limit of the total amount of data processed by the connection. Represents a node i Total limit on the amount of data processed by the connection; (32) The carbon emissions of data centers located in different locations are also uncertain, which in turn affects the carbon emission costs of data centers, as expressed by the following formula: （4） （5） In the formula, EF i Represents a node i Carbon emission factors; p i,renewable Represents a node i The proportion of renewable energy; EF fossil The carbon emission factor representing fossil fuels, EF renewable The carbon emission factor representing renewable energy; V E i It is a data center i Carbon emissions; m i , n i They are nodes i The number of servers and storage devices; p se The operating power of a single server. p st The operating power of a single memory unit. The PUE value for the data center. c ele,i Building nodes for data centers i Electricity price, T It takes one year.

4. The data center site selection and planning method considering decision-dependent uncertainty according to claim 1, characterized in that: Based on the aforementioned decision dependency uncertainty set, a two-stage robust optimization model considering decision dependency uncertainty is constructed as follows: The initial construction and subsequent operation of a data center is a two-stage issue in terms of time scale. The initial construction costs include land costs, construction engineering costs, equipment configuration costs, and communication fiber optic construction costs. The operating costs for the second phase after construction are completed include electricity costs, operation and maintenance costs, carbon emission costs, labor costs, bandwidth and traffic costs for load transfer, and penalty costs incurred when computing power demand cannot be met. The objective function is to minimize the site selection planning cost of the data center considering load shifting capability. In order to more accurately simulate the various costs of data center construction, the planning cost is calculated as the total cost of building the data center for one year. The construction cost is amortized using the equal annual value method. The carbon emission factor is divided into multiple segments within one year for fitting, and it is assumed that each computing power demand node is connected to only one data center. The objective function includes the equivalent annual construction cost and annual operating cost of the data center. The objective function for establishing the two-stage robust model for data center construction and investment is shown below: （6） In the formula: C bu The equivalent annual construction cost of the data center; C op The annual operating cost of the data center; the first phase decision variables include a i and b ij ,in a i Indicates the planned construction location of the data center i Whether to invest or not is a 0-1 variable. b ij For data centers i With computing power demand nodes j Whether the 0-1 variables are connected; d i,ini For nodes i Before the data center shifts its load, the sum of the data received from the connected computing power demand nodes is the main manifestation of decision-related uncertainty. d i For the second-stage decision variables, denoted as nodes i The amount of data that needs to be processed after the data center load is transferred.

5. The data center site selection and planning method considering decision-dependent uncertainty according to claim 4, characterized in that: The equivalent annual construction cost C bu Specifically, it includes: (51) Land costs The land acquisition cost for data center construction nodes is determined by the land transfer price per unit area in the node region and the land area occupied by the data center, as shown in the following formula: （7） In the formula: C L Total land cost for the data center; L p,i The unit price for land transfer. L s,i For the area occupied, The benchmark discount rate for funds; y The number of years for amortizing the investment; I Plan and construct a set of nodes for the data center; a i Indicates the planned construction location of the data center i A 0-1 variable indicating whether or not to invest; (52) Construction costs The construction cost of a data center includes the configuration costs and building installation costs of various systems within the data center, including power supply and distribution systems, cooling systems, and internal communication systems, as shown in the following formula: （8） In the formula: This refers to the total construction cost of the data center. F s,i For nodes i The cost of data center construction; (53) Equipment configuration costs Equipment configuration costs refer to the cost of configuring servers and storage in a data center. The number of servers and storage units is determined by the amount of data the data center needs to process, as shown in the following formula: （9） In the formula: C eq Total equipment configuration cost for the data center; m i , n i They are nodes i The number of servers and storage devices C se and C st These are the unit prices for servers and storage, respectively. (54) Construction costs of optical fiber communication The cost of constructing fiber optic cables for data centers consists of two parts: the cost of constructing fiber optic cables connecting data centers to each other, and the cost of constructing fiber optic cables connecting computing power-demanding nodes to the data center, as shown in the following formula: （10） In the formula: C f This refers to the total cost of constructing fiber optic communication cables for the data center. C opf This refers to the unit cost of constructing optical fiber for communication. D ik This refers to the distance between two data center nodes. D ij The distance between the data center and the nodes requiring computing power; In summary, the formula for the equivalent annual cost of data center construction is as follows: \ MERGEFORMAT (1)。 6. The data center site selection and planning method considering decision-dependent uncertainty according to claim 5, characterized in that: The annual operating cost of the data center C op Specifically, it includes: (61) Electricity costs The servers and storage configured in the data center need to meet the received data center requirements, so they must be kept fully operational. In addition, servers and other IT equipment generate heat, necessitating the operation of cooling systems. The power consumption of the cooling system and other equipment is determined by the data center's power efficiency rating. Therefore, the data center's electricity costs are as follows: （12） In the formula: C e Total electricity cost for the data center; p se The operating power of a single server. p st The operating power of a single memory unit. The PUE value for the data center. c ele,i Building nodes for data centers i Electricity price, T It takes one year; (62) Operation and maintenance costs Data center operation and maintenance costs refer to the expenses incurred in maintaining and servicing various equipment during the operation of the data center. This cost is calculated by multiplying the data center's equipment configuration and construction costs by a fixed percentage, as detailed below: （13） In the formula: C ma The total operating and maintenance cost of the data center, η The percentage of operating and maintenance costs; (63) Carbon emission costs The carbon emission costs of data centers are as follows: （14） （15） In the formula: C co2 It is the total carbon cost of data centers, c co2,i Data center construction node i The unit price of carbon emissions, It is a data center i λ represents the carbon emissions, α represents the base carbon price, and α represents the carbon emission unit price growth rate. l The length of the billing range based on the basic carbon emission unit price, as i As the carbon emissions of data centers at nodes increase, the price per unit of carbon emissions gradually increases. (64) Labor costs The labor costs of a data center are determined by the salary levels and number of employees. Employee salaries include basic salary, social security contributions, and housing provident fund contributions. The salary levels of personnel at different nodes vary and are specifically assessed based on the local GDP level, as shown in the following formula: （16） In the formula: C w It is the total labor cost of the data center. q i It is a node i Number of employees in the data center w i It is a node i The annual income level of employees; (65) Bandwidth and flow costs for load transfer When a data center transfers received data processing requests to other node data centers via fiber optic cables, bandwidth costs are incurred. The bandwidth costs incurred by nodes requesting computing power when transmitting data to the data center are not borne by the data center but by the user. The unit price of bandwidth for data transmission varies between different nodes, as shown in the following formula: （17） In the formula: C bf The total bandwidth cost for the data center. c bf,i For nodes i Data center bandwidth traffic unit price d i,ini For nodes i Before the data center shifts its load, the sum of the data received from the connected computing power demand nodes is as follows: d i This represents the amount of data that needs to be processed after the load is transferred from data center i. (66) Penalty Fees If a data center cannot meet its computing power requirements during operation, it will incur certain penalty costs, as shown in the following formula: （18） In the formula: C pu Penalty fees for data centers that fail to meet demand. d j This indicates the amount of computing power required by the computing power demand node; c pu The penalty cost per unit of computing power lost; In summary, the formula for the annual operating cost of a data center is as follows: （19）。 7. The data center site selection and planning method considering decision-dependent uncertainty according to claim 6, characterized in that: An improved Benders & CCG algorithm is used to solve the two-stage robust optimization model, generating the optimal data center site selection and construction scheme, as follows: (71) Since the maximum value and the multiplication of bilinear terms in formula (16) introduce a nonlinear model, the McCormick envelope is used to linearize the model. The nonlinearity processing strategy is as follows: Since the total bandwidth cost includes a maximum value, a new variable is introduced first. z i Replacing the max term, equation (16) is transformed into the following equation: （20） （21） （22） The McCormick envelope can relax a non-convex problem into a convex problem; therefore, using the McCormick envelope can reduce the bandwidth and flow costs of load transfer. C bf The bilinear terms are transformed into a linear problem, and the final equation (16) is replaced by the following formula: （23） （24） (72) For the two-stage robust problem that takes into account the uncertainty of decision dependency, the improved Benders & CCG algorithm is used to solve it. The improved Benders & CCG algorithm fully considers the uncertainty of decision dependency and introduces three sub-problems into the traditional CCG algorithm. Sub-problem 1 is used to determine whether the original problem has a solution. Sub-problems 2 and 3 are solved iteratively for the two cases of having a solution and not having a solution, respectively. The specific solution steps are as follows: Step 1: Initialization Set LB=-∞, UB=+∞, and set the poles. set With polar rays set It is an empty set; Step 2: Solve the main problem Solve the following main problem: （25） In the formula, Let c1 represent the objective function of the two-stage robust problem, and let c1 represent the coefficient matrix of the decision variables in the first stage. A vector representing the decision variables in the first stage. This represents the objective function value in the second stage. X This represents the set of decision variables for the first stage. This represents the vector composed of poles. This indicates that the poles correspond to constraints where decision-making depends on uncertainty. OU represent x and The KKT conditions, where λ represents the Lagrange multiplier corresponding to the pole in the KKT conditions. Let represent the set of poles, γ represent the vector composed of polar rays, and d-B1x-Ev represent the constraint on the decision-dependent uncertainty corresponding to the polar ray. OV Describe the KKT conditions for v and γ. This represents the Lagrange multiplier corresponding to the polar ray in the KKT condition. Represents a collection of polar rays; Solving for the solution yields the solution. and η ,in η It's a two-phase max-min problem, updating the lower bound. ; Step 3: Solving Subproblem 1 Solving subproblem 1 yields the uncertainty. as well as : （26） In the formula, This represents the objective function value of subproblem 1. Represents uncertain variables. The uncertainty set represents the set of uncertainties that the decision depends on. These represent the decision variables for the second stage. Step 4: About Classification of results if Calculate subproblem 2 and obtain , and the corresponding poles : （27） In the formula, This represents the objective function value of subproblem 2. Let represent the objective function of the second-stage problem, and y be the vector set of decision variables for the second stage. This indicates the constraints for the second phase; renew Substitute the following two constraints into the main problem: （28） （29） if Calculate subproblem 3 to obtain the polar ray. and set : （30） renew Substitute the following two constraints into the main problem: （31） （32） Step 5: Update Step 6: If ,get And terminate the operation; otherwise, t=t+1 and return to step 2 to continue the iteration.

8. A data center site selection planning system that considers decision-dependent uncertainty, used to implement the data center site selection planning method that considers decision-dependent uncertainty as described in any one of claims 1 to 7, characterized in that, include: The building module is used to construct a collaborative planning framework for data centers that takes into account spatiotemporal transfer characteristics; The decision dependency uncertainty set establishment module is used to establish the decision dependency uncertainty set for data center site selection planning, using the data center collaborative planning framework as the physical carrier. The model building module is used to construct a two-stage robust optimization model that considers the uncertainty of decision dependencies based on the decision dependency uncertainty set. The model solution output module uses the improved Benders & CCG algorithm to solve the two-stage robust optimization model and generate the optimal construction plan for data center site selection.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor loads and executes the computer program, it employs the data center site selection planning method that takes into account decision-dependent uncertainty, as described in any one of claims 1 to 7.

Images