Data center energy storage standby battery economic operation method based on multi-stage robust optimization

Abstract

The invention provides a data center energy storage standby battery economic operation method based on multi-stage robust optimization. The method comprises the following steps: S101, setting a dynamic adjustable capacity range of an energy storage standby battery based on a data center standby capacity demand; s102, constructing a multi-stage robust optimization model considering multiple uncertainties of load, electricity price and renewable energy output based on the dynamic adjustable capacity range of the energy storage standby battery; and S103, converting the multi-stage robust optimization model into a two-stage robust optimization model through an affine transformation method, and solving by adopting a nested column and constraint generation algorithm to obtain a data center energy storage real-time optimal scheduling strategy. The operation method provided by the invention improves the operation robustness of the data center energy storage standby battery while ensuring the economical efficiency, and provides a theoretical basis for building a green data center.

Description

Technical Field

[0001] The present invention relates to the field of energy storage technology, and in particular to an economical operation method of a data center energy storage backup battery based on multi-stage robust optimization. Background Art

[0002] With the high-quality development of my country's distribution network and the promotion of the "new energy + energy storage" model, the operating reliability of data centers has been greatly improved. A large number of energy storage backup batteries in data centers have become "sleeping" flexible resources. In order to awaken these flexible resources, it is necessary to formulate reasonable operating strategies based on the negative impact of load, electricity prices, and uncertainty in renewable energy output. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization, so as to solve or at least partially solve the above-mentioned problems existing in the prior art.

[0004] To achieve the above-mentioned object of the invention, the present invention provides a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization, the method comprising the following steps:

[0005] S101. Setting a dynamically adjustable capacity range of the energy storage backup battery based on the backup capacity requirements of the data center;

[0006] S102. Construct a multi-stage robust optimization model that considers multiple uncertainties of load, electricity price, and renewable energy output based on the dynamically adjustable capacity range of energy storage backup batteries;

[0007] S103. The multi-stage robust optimization model is transformed into a two-stage robust optimization model through an affine transformation method, and a nested column and constraint generation algorithm is used to solve it, thereby obtaining a real-time optimal scheduling strategy for data center energy storage.

[0008] Furthermore, the upper and lower limit constraints of the state of charge based on the energy storage backup battery equipment constraints are expressed as:

[0009]

[0010] In the formula, is the actual charge state of the energy storage backup battery, act is the real-time charge state when only the battery equipment constraints are considered, UPS is the energy storage battery of the data center, t is the time, is the lower limit constraint of the charge state of the energy storage backup battery, It is the upper limit constraint of the state of charge of the energy storage backup battery.

[0011] Furthermore, There is a linear relationship with the load of the next period in the system, that is, It depends on the backup demand in period t+1, which is expressed as:

[0012]

[0013] In the formula, Data center energy storage backup battery factor for data center operation standard, Soc UPS,t+1 is the charge state of the data center at time t+1.

[0014] Furthermore, the target cost of the multi-stage robust optimization model of the data center energy storage system considering the uncertainty of both the source and the load and the electricity price is expressed as:

[0015]

[0016] In the formula, min x is the outer function, is the inner max model decision function of the first period, is the inner min model function, is the inner max function, is the inner min model function, is the standard form of robust optimization in the first period, Q2(y1) is the worst function of y1 in the second period based on the first period, Q t (y t-1 ) is the (y of period t based on period (t-1) t-1 ), Q t+1 (y t ) is the (t+1) period based on the (y t ), is the standard form for robust optimization, is the expected value of the total electricity purchase cost of the data center energy storage system and the upper power grid under the worst distribution of electricity purchase prices, is the expected value of the total electricity sales profit of the data center energy storage system and the upper power grid under the worst electricity sales price distribution, represents the charge and discharge deterioration cost of ESS during the scheduling period, x is the set of decision variables of the min model of the outer main problem, is the electricity sales status at time t, is the electricity purchasing status at time t, u t is the decision variable of the inner max model, is the actual power of the load at time t, is the actual power of wind power at time t, y t is the set of decision variables of the inner min model, The power purchasing strategy for the data center is determined based on the overall system demand and external electricity prices. The power sales strategy for the data center is determined based on the overall system demand and external electricity prices. is a 0-1 variable, is a 0-1 variable, The charging power of the data center energy storage system, is the discharge power of the data center energy storage system, SOC t Is the battery charge state.

[0017] Furthermore, the affine transformation form of converting the multi-stage robust optimization model into the two-stage robust optimization model is:

[0018]

[0019] In the formula, The charging power of the data center energy storage system, is the discharge power of the data center energy storage system, Soc UPs,t is the charge state of the data center at time t, is the actual power of wind power at time t, is the actual power of the load at time t, The decision variables optimized for the main problem, also known as the affine function constant matrix.

[0020] Furthermore, after affine transformation, the target cost of the multi-stage robust optimization model of the data center energy storage system is transformed from a multi-stage robust optimization model to a two-stage robust optimization model, taking into account the uncertainty of both the source and the load and the electricity price, including:

[0021] min x {A T x+max u min y B T y}

[0022]

[0023] In the formula, min x is the outer function, max u is the inner max function, min y is the inner min function, A T x and B T y is the standard form of the "abstract form" of two-stage robust optimization, x is the set of decision variables of the outer main problem min model, u is the decision variables of the inner max model, and y is the set of decision variables of the inner min model. is the electricity sales status at time t, is the electricity purchasing status at time t, The power purchasing strategy for the data center is determined based on the overall system demand and external electricity prices. The power sales strategy for the data center is determined based on the overall system demand and external electricity prices. is a 0-1 variable, is a 0-1 variable, is the distributed blue stick multiplier of the electricity purchase variable, is the distributed blue stick multiplier of the electricity sales variable.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention proposes an economic operation method for energy storage backup batteries in data centers based on multi-stage robust optimization. First, the dynamically adjustable capacity range of energy storage backup batteries is set based on the backup capacity demand of the data center. Secondly, a multi-stage robust optimization model is constructed that takes into account multiple uncertainties of load, electricity price, and renewable energy output; then, the multi-stage robust optimization model is converted into a two-stage robust optimization model through affine transformation, and a nested column and constraint generation algorithm is used to solve it. The solution results show that the proposed operation method improves the operation robustness of the energy storage backup batteries in the data center while ensuring economy, providing a theoretical basis for building a green data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 It is a step diagram of the economic operation method of data center energy storage backup battery for multi-stage robust optimization;

[0028] Figure 2 It is the energy storage lower limit correction curve diagram;

[0029] Figure 3 It is the system architecture design diagram;

[0030] Figure 4 It is a schematic diagram of the solution process;

[0031] Figure 5 It is a sample curve chart of electricity price;

[0032] Figure 6 This is a PV output uncertainty analysis diagram based on DRCC;

[0033] Figure 7 It is the worst wind power and charge power diagram considering different robust parameters;

[0034] Figure 8 It is a multi-stage robust optimization scheduling strategy graph. DETAILED DESCRIPTION

[0035] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0036] Example

[0037] This embodiment provides a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization, the method comprising the following steps:

[0038] Figure 1 Diagram of steps for economic operation of data center energy storage backup batteries for multi-stage robust optimization.

[0039] S101. Setting a dynamically adjustable capacity range of the energy storage backup battery based on the backup capacity requirements of the data center, specifically including:

[0040] The upper and lower limits of the state of charge (SOC) based on the energy storage backup battery equipment constraints are:

[0041]

[0042] In the formula, It represents the real-time charge state of the energy storage backup battery, act represents the real-time charge state when only the battery equipment constraints are considered, and UPS represents the data center energy storage battery. and It is a fixed value, but in actual situations, if we only consider the upper and lower limits of the fixed value, it may not be able to meet the complex conditions in actual operation, that is, when the data center has a high load rate and a corresponding high backup demand, the energy storage battery is On the contrary, when the data center load rate is low and the backup demand is reduced, the energy storage battery It may be too low, causing battery over-discharge problems.

[0043] To this end, the present invention After modification, a method for calculating the dynamic dispatchable capacity of energy storage backup batteries in data centers is proposed. There is a linear relationship with the load of the next period in the system, that is, It depends on the backup demand in period t+1, which is expressed as:

[0044]

[0045] In the formula, The data center energy storage backup battery coefficient, Soc UPS,t+1 It represents the charge state of the data center at time t+1.

[0046] The data center energy storage backup battery factor for data center operation standards is:

[0047]

[0048] In the formula, It is expressed as the upper limit of load demand based on uncertainty in the next period, It is expressed as the minimum backup time of the energy storage backup battery at time t, E rated Expressed as the rated capacity of the data center energy storage battery.

[0049] When the communication load is low, that is, Soc UPS,t+1 If the value is too small, the SOC lower limit may be too low, resulting in over-discharge and endangering the safety and reliability of system operation. Lower limit Corrected to a piecewise function:

[0050]

[0051] In the formula, Soc UPS,24 It represents the charge state of the data center energy storage at 24 points, with a fixed value of 0.5.

[0052] Figure 2 (a) is the lower limit of the SOC of the energy storage backup battery considering only the energy storage device constraint or the communication load rate; Figure 2 (b) is the lower limit of SOC of the energy storage backup battery that takes into account both energy storage device constraints and backup requirements.

[0053] In order to evaluate the regulation capability of the data center energy storage equipment, the present invention first subtracts the backup demand under different load conditions from the configuration capacity of the data center energy storage backup battery, thereby obtaining the remaining adjustable capacity and charge and discharge power of the data center energy storage backup battery.

[0054] like Figure 2 As shown in (b), the backup energy storage capacity of the data center is expressed as:

[0055]

[0056] In the formula, E UPS,t represents the spare capacity of the data center under the communication load condition during period t, It is expressed as the lower limit of the dischargeable capacity when self-provided energy storage is used in period t; the adjustable range of the data center energy storage is obtained from the upper and lower limits of the SOC of the data center energy storage battery.

[0057] The maximum values ​​of the chargeable capacity and dischargeable capacity of the data center's self-provided energy storage in period t are:

[0058]

[0059] In the formula, It is expressed as the maximum value of the rechargeable capacity of the data center's self-provided energy storage during period t, It is represented by the maximum discharge capacity of the data center's self-provided energy storage during period t, ch represents charging, dch represents discharging, and Soc UPS,t It is represented by the SOC value of the data center in period t, E max It represents the upper limit of the charging depth allowed for the data center’s own energy storage.

[0060] S102. Construct a multi-stage robust optimization model based on the dynamically adjustable capacity range of the energy storage backup battery, taking into account the multiple uncertainties of load, electricity price, and renewable energy output, including:

[0061] Considering the uncertainty of renewable energy generation, the data center power system is modeled. In order to achieve friendly coordinated interaction between data center energy storage and the power grid, the objective function is defined as:

[0062]

[0063]

[0064] Where, formula (7) represents the target cost of the multi-stage robust optimization model of the data center energy storage system considering the source-load dual side and electricity price uncertainty. x is represented as an outer function, It is represented as the inner max model decision function of the first period, Represented as the inner min model function, Represented as the inner max function, Represented as the inner min model function, It is expressed as the standard form of robust optimization in the first period, Q2(y1) is expressed as the worst function of y1 in the second period based on the first period, Q t (y t-1 ) is expressed as (y in period t based on period (t-1) t-1 ), Q t+1 (y t ) is expressed as (t+1) period based on (y t ), Expressed in standard form for robust optimization, and They represent the total electricity purchase cost and expected value of electricity sales profit of the data center energy storage system and the upper power grid under the worst distribution of electricity purchase and sales prices, represents the charging and discharging deterioration cost of ESS during the scheduling period, and They represent the electricity purchase / sale status at time t, and They represent the power purchase / sale strategies of the data center based on the overall system demand and external electricity prices. and Respectively represent the actual power of wind power and load at time t, and is a 0-1 variable, E P represents the expected value of the cost of purchasing and selling electricity, and Indicates the electricity price, soc t Indicates the battery charge state, N (buy)sell represents the number of samples of random (buy) and ask prices, C E represents the cost unit coefficient, and They represent the charging and discharging power of the data center energy storage system, x represents the set of decision variables of the outer main problem min model, and u t Represented as the inner max model decision variable, y t Represented as the set of decision variables of the inner min model.

[0065] Expressed in the form of Wasserstein metric and The uncertain set:

[0066]

[0067] In the formula, represents the fuzzy set of electricity purchase / sale price at time t, represents the empirical distribution of electricity purchase / sale prices at time t, and They represent the random variables of the purchase / sale price of electricity from the data center to the upper power grid, P represents the set of probability distributions of the purchase / sale price, and They represent the empirical distribution of the purchase / sale electricity price at time t, and ε represents the distribution P and the empirical distribution The distance between them.

[0068] Distribution P and empirical distribution The distance between is defined as:

[0069]

[0070] In the formula, |ξk -ξ i || p represents the p-norm, which is usually taken as 1 in calculations. K and N represent the distribution P and The total number of samples, P kI represents the joint distribution (p, ) Take a sample (ξ k ,ξ i )’s probability size.

[0071] Modeling considering multiple uncertainties: Considering two different renewable energy access schemes, namely, a wind power generation system with external centralized power generation and a rooftop photovoltaic system with internal distributed power generation, the system architecture design of RES+ data center participating in the economic dispatch of the power grid is as follows: Figure 3 shown.

[0072] For data center systems, the load demand and external wind power generation output volatility can be modeled using equation (14) for uncertainty.

[0073]

[0074] In the formula, and They represent the predicted values ​​of wind power and load at time t, WD represents wind power, Load represents load power, and pre represents the predicted value (reference value). and They represent the actual power of wind power and load at time t, ζ represents the fluctuation range of wind power and load uncertainty, as well as All are expressed as 0-1 variables, W represents wind power, L represents load power, UB represents the upper bound, UB = 1 when the power is maximum, LB represents the lower bound, LB = 1 when the power is minimum, Γ WD and Γ Load It represents the robust uncertainty of uncertain parameters. The larger the corresponding value is, the stronger the uncertainty of the parameter is, and the more conservative the corresponding decision is.

[0075] Modeling of distributed photovoltaic output and its uncertainty: The distributed robust chance constraint based on Wasserstein distance is introduced to represent the power balance constraint of the system, and its expression is shown in formula (15).

[0076]

[0077] Where P pv represents the photovoltaic output distribution set in the sample set, The random variable representing the photovoltaic output at time t.

[0078] Formula (15) can be directly transformed into the following as the distributed robustness constraint by the boundary support method:

[0079]

[0080] The above transformation process, ε pv represents the radius of the Wasserstein sphere, and is the corresponding auxiliary variable, N pv represents the number of PV samples at each moment, β represents the violation probability, represents the upper boundary of the historical sample photovoltaic output at time t, represents the photovoltaic output of the Ith sample at time t in the sample set, pv represents photovoltaic output, I represents the Ith sample, Represents the lower boundary of the historical sample photovoltaic output at time t.

[0081] Formulas (16-19) obtained by the boundary support method still have problems: Formula (17) contains many bilinear terms, which makes the transformed model highly non-convex; when the number of samples N pv If the value is large, the total number of equations (17-19) will become particularly large, resulting in a large amount of computing memory consumption. In order to solve these problems, equation (15) is reconstructed as follows:

[0082]

[0083] Among them, P pv Represented as the photovoltaic output distribution set in the sample set, It is expressed as the actual output at time t, The random variable of photovoltaic power at time t is represented by the variable on the left side of formula (20) Nested on the right side of formula (21):

[0084]

[0085] Under the premise of probability 1-β, the larger Therefore, in order to obtain a boundary that is more in line with the actual photovoltaic output Formula (20) can be used as the internal constraint of the following optimization problem:

[0086]

[0087] Solve equation (22) and use the boundary support method to make equation (20) equivalent:

[0088]

[0089] The optimal solution of formula (23) yes Satisfies the strict lower bound of the probability constraint. Once it is solved, Substitute it directly into the right side of equation (21) and regard it as the power balance constraint inside the data center as equation (24).

[0090]

[0091] In this embodiment, the energy storage control constraints include data center energy storage system operation constraints, power purchase and sales constraints, and data center power balance constraints.

[0092] The operating constraints of the data center energy storage system are:

[0093]

[0094] Soc UPS,init =Soc UPS,24 (29)

[0095]

[0096] In the formula, They represent the maximum charging and discharging power of the current energy storage, and represents a 0-1 variable, It represents the lower limit of SOC of the data center energy storage battery at time t; represents the SOC upper limit of the data center energy storage battery at time t, Soc UPS,t It represents the state of charge of the data center at time t, Soc UPS,init Indicates the initial charge state of the data center energy storage, init indicates the initial value, η ch and η dch Respectively represent the charging and discharging efficiency of energy storage.

[0097] The power constraints for purchasing and selling electricity are:

[0098]

[0099] In the formula, and They are the upper limits of power purchase and power sales respectively; and are 0-1 variables, representing the electricity purchasing and selling status at time t.

[0100] The power balance constraint of the data center is:

[0101]

[0102] In the formula, is the renewable wind power generation at time t, is the internal load demand of the data center at time t, The distributed photovoltaic power generation of the data center.

[0103] S103, transforming the multi-stage robust optimization model into a two-stage robust optimization model by using an affine transformation method, and solving it by using a nested column and constraint generation algorithm to obtain a real-time optimal scheduling strategy for energy storage in a data center, specifically including:

[0104] Solve the multi-stage robust model:

[0105] The multi-stage robust optimization model solution process is as follows Figure 4 shown.

[0106] Reconstruction of J1 and J2: Equation (8-9) is equivalently transformed into (33) to (35):

[0107]

[0108] In the formula, ε pb and ε ps are the Wasserstein radius of the electricity purchase and sale price, Distribution represents the distribution of the Bruno multiplier of the purchase / sale electricity variable, and represents the electricity purchase and sales prices at time t on the Ith day in the electricity price sample set, I represents the Ith day, ps represents sales, pb represents purchase, and N buy and N sell Respectively represent the number of samples of purchase and sale electricity prices.

[0109] Compared with equations (8-9), equations (33-35) transform the "min-max" distributed terms related to the purchase / sale electricity price in each time period into a single-layer "min-min" optimization, which greatly simplifies the model.

[0110] Refactor J3:

[0111] J3 represents the charging and discharging deterioration cost of data center energy storage, which includes three decision variables: charging, discharging, and state of charge. The results of the three in the current period will directly affect the next period, so they need to be equivalent with the help of affine transformation.

[0112] The state quantities involved in affine transformation mainly include Soc UPS,t , the specific affine transformation form is as follows:

[0113]

[0114] In the formula, It is represented by the charging power of the data center energy storage system, It is expressed as the discharge power of the data center energy storage system, Soc UPs，,t It is represented as the charge state of the data center at time t, It is expressed as the actual power of wind power at time t, It is expressed as the actual power of the load at time t, It represents the decision variables optimized for the main problem, also known as the affine function constant matrix. Once the main problem is decided, it is fixed.

[0115] Through affine transformation, the deterioration cost of energy storage can be equivalently changed to:

[0116]

[0117] At the same time, the constraints of equation (25-30) on charging and discharging are also equivalent to:

[0118]

[0119] After affine transformation, the objective function is transformed from a multi-stage robust optimization model to a two-stage robust optimization:

[0120]

[0121] In the formula, min x Represented as an outer function, max u Represented as inner max function, min y is represented as the inner min function, x is represented as the set of decision variables of the outer main problem min model, u is represented as the decision variable of the inner max model, and Y is represented as the set of decision variables of the inner min model. Since the inner min model contains 0-1 variables, It is expressed as the distributed blue stick multiplier of the electricity purchase variable, The distributed Robust multiplier of the electricity sales variable is expressed as the nested column and constraint generation algorithm is used to split the original problem into a series of problems for solution.

[0122] The solution is solved using the Nested Column and Constraint Generation (Nest-CCG) algorithm, including:

[0123] The data center optimization model based on affine transformation is:

[0124]

[0125] Using the Nest-CCG algorithm to solve it, it is necessary to split formula (A1) into the main problem and sub-problems for iterative solution. In the kth iteration, the master problem (MP) corresponding to formula (A1) is:

[0126]

[0127] The corresponding constraints are:

[0128]

[0129] In (A3-A14), Hehe It is represented as the lth actual scenario of wind power and load in the kth iteration, which is mainly obtained by solving the sub-problems.

[0130] In the kth iteration, since the "max-min" problem corresponding to equation (A1) contains 0-1 variables, it cannot be directly transformed using the KKT condition. It needs to be split into a sub-problem of the sub-problem (SPS) and a main problem of the sub-problem (SPM) for solution. When the sub-problem converges internally, the corresponding wind power and load uncertainty set scenarios are output to the main problem.

[0131] The SPS model of the data center is:

[0132]

[0133] The constraints are:

[0134]

[0135] In formula (A16-A22), and is the optimal solution of the MP problem in the kth iteration, and and When the main problem is iterated for the kth time, the optimal solution of SPM obtained in the Vth iteration of the subproblem, v represents the number of iterations of the subproblem, and Represented as wind power and load power in the Vth subproblem iteration.

[0136] (3) The SPM model of the data center is:

[0137] The maxθ(A23) constraint is:

[0138]

[0139]

[0140] In formula (A24-A47), and is the optimal solution of the MP problem in the kth iteration, and and The optimal solution obtained by the vth SPS in the Vth iteration of the subproblem at the kth iteration of the main problem. D is the uncertainty set in formula (14) of the main text.

[0141] Experimental example

[0142] This experimental example provides an economic operation method for a data center energy storage backup battery based on multi-stage robust optimization. This experimental example performs case analysis according to the above method, and the case analysis includes:

[0143] 1. Basic data and scene settings:

[0144] (1) Example The model and algorithm of this paper are verified by taking wind power, photovoltaic power, data center and load aggregation as examples. The energy storage parameter settings of the data center are shown in Table 1:

[0145] Table 1 Simulation parameter settings

[0146]

[0147]

[0148] Since the interactive electricity price between the data center and the upper power grid is uncertain, the present invention samples 50 days of real-time purchase / sale electricity prices. The corresponding sample curves and average electricity prices are shown in Figure 2. Figure 5 shown.

[0149] (2) The photovoltaic data comes from an actual photovoltaic power station, the load data comes from the literature, and the real-time electricity price samples of the example sampling are all from the PSEG region and sampled from the PJM official website. pb and ε ps In this example, the Wasserstein radius used in the same scene is the same, and is uniformly represented by ε. The convergence accuracy of the Nest-CCG algorithm is eps=10 -6 The complementary relaxation large M variable is set to 10000. The compiler software used is Matlab2023a, and the commercial solver is CPLEX 12.10.0. The iterative process corresponding to the algorithm is shown in Table 2.

[0150] Table 2 Iterative solution based on Nest-CCG

[0151]

[0152]

[0153] (3) The present invention samples the photovoltaic power inside the data center for 50 days, and the corresponding Wasserstein radius ε pv In order to obtain the lower bound of photovoltaic output that meets the distributed robustness constraint, the 50-day photovoltaic historical data is substituted into formula (23), and the lower bound of photovoltaic output from 9:00 to 18:00 is calculated by combining different confidence levels. The solver used is Gurobi9.1.2, and the final optimization result is as follows Figure 6 shown.

[0154] from Figure 6 It can be seen that the lower boundary output of photovoltaic increases with the increase of β, which is also in line with the theory of probability quantiles, that is, the larger the quantile, the smaller the value of the random variable. When the value of β is lower than 0.1, the lower boundary of photovoltaic output is smaller than most scenarios. Therefore, the parameter "illegal probability" β in formula (19) should not be too small, otherwise the optimization result of formula (23) will be more conservative. Combined with the photovoltaic boundary output curve, this paper sets the "illegal probability" β in formula (15) to 0.2.

[0155] (4) Based on the combination of all the above uncertainties, this example sets the following scenario:

[0156] Scenario 1: Sampling days and Wasserstein radius for different electricity prices;

[0157] Scenario 2: Different load uncertainties (Γ Load )

[0158] Scenario 3 Different distributed power output (Γ WD ) Comparison under uncertainty;

[0159] Scenario 4 considers the comparison under different loads and distributed power output uncertainty.

[0160] 2. Analysis of simulation scenario results

[0161] When the lower limit of the system battery energy storage does not consider the communication load rate, but is based only on the equipment constraint (0.1), the optimized operation result will result in the lower limit of the energy storage being 0.1 during peak hours, such as 20:00 and 21:00. This cannot meet the backup requirements of the communication load and is not conducive to the stability and reliability of the system. Therefore, it is very necessary to comprehensively consider the energy storage equipment constraints and backup requirements, and establish a more reasonable lower limit of the energy storage backup battery SOC so that the system always has sufficient backup capacity to cope with potential load fluctuations.

[0162] The parameter settings of each scenario are shown in Table 2, and the simulation results are shown in Table 3. In order to take into account the uncertainty of electricity prices as much as possible, the number of sample days for the average electricity price of scenarios 2, 3, and 4 is 50. For the convenience of comparison, Γ in scenario 1 Load and Γ WD Fixed to 8, compared with different ε pb / ε ps and the optimal dispatch cost under the average electricity price; Γ in scenario 2 Load Fixed to 8, comparing different ε and Γ WDThe optimal scheduling cost under Γ in scenario 3 WD Fixed to 8, comparing different Γ Load Impact on scheduling cost. In scenario 4, ε is fixed to 0.02 and Γ is changed Load and Γ WD The corresponding worst wind power output and load demand data are as follows: Figure 7 As shown, in Figure 7 (a) is the worst wind power considering different robust parameters, and (b) is the worst load power considering different robust parameters.

[0163] Table 3 Simulation parameter settings

[0164]

[0165]

[0166] According to Table 3, when other parameters remain unchanged, when ε increases, the scheduling cost of the system increases accordingly. Load When Γ increases, the robustness of the system increases, while the scheduling cost only increases by 6.2%, 3.4%, and 2.5%; in scenario 3, when other parameters remain unchanged, WD When increases, the robustness of the system increases, while the dispatch cost only increases by 7.2%, 5.1%, and 3.7%; in scenario 4, considering the uncertainty of different loads and distributed power output, the dispatch cost increases by a maximum of 26.7%. At this time, although the cost increases, the robustness of the system is improved.

[0167] Table 4 Comparison of optimal scheduling costs in various scenarios

[0168] Table 4.1 Comparison of optimal scheduling costs in scenario 1

[0169]

[0170] Table 4.2 Comparison of optimal scheduling costs in scenario 2

[0171]

[0172] Table 4.3 Comparison of optimal scheduling costs in scenario 3

[0173]

[0174] Table 4.4 Comparison of optimal scheduling costs for scenario 4

[0175]

[0176] 3. Analysis of real-time multi-stage optimization scheduling results of data center energy storage

[0177] The real-time optimal scheduling strategy for data center energy storage is obtained by solving the optimization model proposed in this invention based on affine transformation combined with Nest-CCG algorithm. Considering different values ​​of ε and the sampled electricity price samples of 30 days, 40 days, and 50 days, the optimal scheduling cost is shown in Table 3.1. When ε increases, the scheduling cost of the system increases accordingly.

[0178] Taking ε = 0.02 and the electricity price sample for 50 days as the total sample size as an example, the real-time optimal scheduling strategy for data center energy storage obtained by the optimization model proposed in the present invention is as follows: Figure 8 As shown, in Figure 8 (a) is the worst scenario for wind power output, (b) is the worst scenario for load demand, (c) is the optimal decision plan for purchasing and selling electricity, and (d) is the optimal scheduling plan for energy storage.

[0179] from Figure 8 From the simulation calculations of (a) to 8(d), it can be seen that during the low electricity price period from 22:00 to 5:00 the next day, the load demand of the prosumer is small, the photovoltaic power generation is basically not working, and the wind power generation is large. The load of the prosumer is fully supplied by the wind turbine, and the energy storage device is charged to absorb the excess wind power output, and finally the remaining electricity is sold in the electricity market;

[0180] During the first peak electricity price period from 7:00 to 9:00 in the daytime, the load demand of prosumers is relatively small and the wind power output is slightly reduced, but the photovoltaic output is in the off-peak period. Considering the minimization of dispatching costs, wind power, photovoltaic and data center energy storage systems cooperate to discharge to meet the load demand, and finally sell the redundant electricity in the power market;

[0181] During the second peak period of electricity prices from 16:00 to 22:00 during the day, the load demand of prosumers is large, while the output of wind power and photovoltaic power is relatively small. The load of prosumers is first supplied by wind power, photovoltaic units and energy storage. The energy storage system of the data center discharges to meet the load demand, and finally purchases the insufficient electricity in the electricity market.

[0182] During the off-peak period between 5:00 and 7:00 a.m., the load demand of prosumers is small, the wind power output is also small, and the photovoltaic power generation is not working. The load of prosumers is supplied by wind turbines and data center energy storage systems. The data center energy storage system discharges to meet the load demand first, and can sell the surplus electricity in the power market to obtain economic benefits.

[0183] During the off-peak period of electricity prices from 9:00 to 16:00 during the day, the load demand of prosumers is moderate, and the wind power output is also small, but the photovoltaic output is at its peak. The load of prosumers is supplied by wind turbines and photovoltaics, and electricity is purchased in the electricity market to charge the data center energy storage system. During the off-peak period of electricity prices, the data center energy storage system is used to store electricity, and during the next peak period of electricity prices, the data center energy storage system is used to release electricity, thereby selling this part of electricity in the electricity market to obtain economic benefits. Therefore, prosumers can reduce overall operating costs by optimizing the scheduling of energy storage and using the price difference for arbitrage, which also helps promote the consumption of renewable energy.

[0184] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for economic operation of energy storage backup batteries in data centers based on multi-stage robust optimization, characterized in that: The method comprises the following steps: S101. Setting a dynamically adjustable capacity range of the energy storage backup battery based on the backup capacity requirements of the data center; S102. Construct a multi-stage robust optimization model that considers multiple uncertainties of load, electricity price, and renewable energy output based on the dynamically adjustable capacity range of energy storage backup batteries; S103. The multi-stage robust optimization model is transformed into a two-stage robust optimization model through an affine transformation method, and a nested column and constraint generation algorithm is used to solve it, thereby obtaining a real-time optimal scheduling strategy for data center energy storage.

2. According to the method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization according to claim 1, in step S101, the dynamic adjustable capacity range of the energy storage backup battery is set based on the backup capacity demand of the data center, including: The upper and lower limit constraints of the state of charge based on the energy storage backup battery equipment constraints are expressed as: In the formula, is the actual charge state of the energy storage backup battery, act is the real-time charge state when only the battery equipment constraints are considered, UPS is the energy storage battery of the data center, t is the time, is the lower limit constraint of the charge state of the energy storage backup battery, It is the upper limit constraint of the state of charge of the energy storage backup battery.

3. According to claim 2, a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization is characterized in that: The upper limit constraint of the state of charge of the energy storage backup battery is set by the dynamic dispatchable capacity calculation method of the data center energy storage backup battery Make corrections, specifically: There is a linear relationship with the load of the next period in the system, that is, It depends on the backup demand in period t+1, which is expressed as: In the formula, Data center energy storage backup battery factor for data center operation standard, Soc UPS,t+1 is the charge state of the data center at time t+1.

4. According to claim 1, a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization is characterized in that: In step S102, it includes: Considering the uncertainty of both the source and the load and the electricity price, the target cost of the multi-stage robust optimization model of the data center energy storage system is expressed as: In the formula, min x is the outer function, is the inner max model decision function of the first period, is the inner min model function, is the inner max function, is the inner min model function, is the standard form of robust optimization in the first period, Q2(y1) is the worst function of y1 in the second period based on the first period, Q t (y t-1 ) is the (y of period t based on period (t-1) t-1 ), is the standard form of robust optimization, Q t+1 (y t ) is the (t+1) period based on the (y t ), is the expected value of the total electricity purchase cost of the data center energy storage system and the upper power grid under the worst electricity purchase price distribution, is the expected value of the total electricity sales profit of the data center energy storage system and the upper power grid under the worst electricity sales price distribution, represents the charge and discharge deterioration cost of ESS during the scheduling period, x is the set of decision variables of the min model of the outer main problem, is the electricity sales status at time t, is the electricity purchasing status at time t, u t is the decision variable of the inner max model, is the actual power of the load at time t, is the actual power of wind power at time t, y t is the set of decision variables of the inner min model, The power purchasing strategy for the data center is determined based on the overall system demand and external electricity prices. The power sales strategy for the data center is determined based on the overall system demand and external electricity prices. is a 0-1 variable, is a 0-1 variable, The charging power of the data center energy storage system, is the discharge power of the data center energy storage system, SOC t Is the battery charge state.

5. According to claim 4, a method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization is characterized in that: The affine transformation form of converting the multi-stage robust optimization model into a two-stage robust optimization model is: In the formula, The charging power of the data center energy storage system, is the discharge power of the data center energy storage system, Soc UPS,t is the charge state of the data center at time t, is the actual power of wind power at time t, is the actual power of the load at time t, The decision variables optimized for the main problem, also known as the affine function constant matrix.

6. The method for economic operation of a data center energy storage backup battery based on multi-stage robust optimization according to claim 5 is characterized in that: After affine transformation, the target cost of the multi-stage robust optimization model of the data center energy storage system is transformed from a multi-stage robust optimization model to a two-stage robust optimization model, taking into account the uncertainty of the source and load sides and the electricity price, including: min x {HAS T x+max u min y B T y} In the formula, min x is the outer function, max u is the inner max function, min y is the inner min function, A T x and B T y is the standard form of the "abstract form" of two-stage robust optimization, x is the set of decision variables of the outer main problem min model, u is the decision variables of the inner max model, and y is the set of decision variables of the inner min model. is the electricity sales status at time t, is the electricity purchasing status at time t, The power purchasing strategy for the data center is determined based on the overall system demand and external electricity prices. The power sales strategy for the data center is determined based on the overall system demand and external electricity prices. is a 0-1 variable, is a 0-1 variable, is the distributed blue stick multiplier of the electricity purchase variable, is the distributed Robust multiplier of the electricity sales variable.

Images