Cooperative operation method for data center cluster and shared energy storage

Abstract

The invention discloses a collaborative operation method for a data center cluster and shared energy storage, and aims to solve the problems of disjunction of long-term planning and short-term operation, difficulty in multi-agent collaboration, conservative uncertainty response, insufficient service quality guarantee and the like of a green power direct connection data center cluster in energy management. The invention provides a two-stage double-layer collaborative management framework. In the first stage, a robust optimization model driven by data is constructed, and the capacity of the shared energy storage and the capacity of the gas turbine are optimally configured. And in the second stage, an asymmetric Nash bargaining game model is constructed, and distributed solution is carried out by using an alternating direction multiplier method so as to realize multi-agent cooperative operation and fair income distribution. According to the method, the cross-time-scale coupling decision problem is effectively solved, synergistic interaction among multiple subjects is realized, and the overall economy and the renewable energy consumption rate of the system are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of energy system engineering and information technology, and particularly relates to a method for cooperative operation of a data center cluster and shared energy storage. BACKGROUND

[0002] Under the background of rapid development of information technology, the energy consumption problem of data centers as the core infrastructure supporting the development of digital economy is increasingly prominent. In order to reduce carbon and improve efficiency, renewable energy such as solar energy and wind energy is gradually introduced into the power supply system of data centers, but the inherent intermittency and volatility of renewable energy and the strict requirements of data centers on power supply stability and reliability form a fundamental contradiction. Especially under the trend of clustering of data centers, the cluster system composed of multiple data centers has higher requirements for the stability of energy supply, which makes the large-scale application of renewable energy face greater challenges.

[0003] The current renewable energy power supply solution for data center clusters mainly has three technical bottlenecks: first, the existing system often adopts a mode of separating offline planning from online operation, which makes it difficult for energy scheduling solutions to adapt to real-time fluctuations in renewable energy output; second, the cooperative scheduling mechanism among multiple data center clusters is not perfect, and there is a lack of effective benefit distribution and risk sharing mechanism. These problems seriously restrict the large-scale application of renewable energy in the field of data centers, and there is an urgent need for a new energy management technical solution that can realize integrated planning and operation, multi-agent cooperative optimization, and robustness. SUMMARY

[0004] Therefore, the application provides a method for cooperative operation of a data center cluster and shared energy storage, aiming to solve the problem in the prior art that the planning and operation scheme of a data center cluster cannot be effectively combined, the uncertainty of renewable energy and load is not reliably handled, or the calculation speed is slow.

[0005] To this end, the application provides the following technical solutions: The application provides a method for cooperative operation of a data center cluster and shared energy storage, which configures shared energy storage to realize distributed optimization configuration for the operating cost of a heterogeneous data center cluster, including the following steps: Collecting historical data, including the output of each renewable energy and the work load of each data center; Based on the historical data, a data-driven robust capacity configuration model is constructed to optimize the capacity of shared energy storage and gas turbines; the objective of the robust capacity configuration model is to minimize the equipment investment cost and the expected operating cost under uncertainty scenarios; A multi-entity collaborative operation model is established based on the optimized configuration of the data center cluster and the shared energy storage. The goal of the multi-entity collaborative operation model is to minimize the total operating cost of the data center cluster and maximize the operating benefits of the shared energy storage system. A distributed optimization algorithm is used to solve the multi-entity collaborative operation model to determine the optimal operation strategy of each data center and the shared energy storage system during the scheduling cycle.

[0006] Furthermore, the multi-entity collaborative operation model is based on an asymmetric Nash bargaining game model to allocate the cooperative benefits generated during collaborative operation; the bargaining power of each entity is calculated according to the contribution quantification function to determine the final benefit distribution scheme of each entity.

[0007] Furthermore, based on the historical data, a data-driven robust capacity configuration model is constructed, including: A data-driven sample robust optimization method is used to process historical data, including: using the historical data to construct an uncertainty set that can represent possible future scenarios; ensuring that the decision is feasible in historical scenarios at least at that confidence level by introducing a confidence level parameter; the confidence level parameter is used to adjust the proportion of samples included in the uncertainty set; Establish a target function for the coordinated configuration of data centers and shared energy storage, wherein the target function is: ; in, To share the energy storage lifecycle conversion factor; Investment cost per unit energy capacity for shared energy storage; The maximum energy capacity for shared energy storage; This is a gas turbine life cycle conversion factor. This refers to the investment cost per unit power capacity of the gas turbine. For data centers i The maximum power capacity of a gas turbine; For the power purchased; Time-of-use pricing for electricity purchases; This refers to the power output for electricity sales. Time-of-use pricing for electricity; This is the energy storage degradation cost coefficient; Total power of shared energy storage charging; To share the total discharge power of the energy storage; The penalty cost coefficient for abandoning scenic views; Solar power curtailment; Wind power curtailment; Natural gas purchase price; Gas turbine power consumption; Cost coefficient of gas turbine refrigeration; Penalty coefficient of unfinished task; Task completion status flag variable; Establish the constraint condition of data center and shared energy storage collaborative configuration, which includes: The gas turbine operation constraint condition is: ; In the formula The power generation power of the gas turbine; The maximum power generation power of the gas turbine; The thermal power of the gas turbine; The thermal-electric ratio coefficient of the gas turbine; The electric refrigerator operation constraint condition is: ; Wherein, The electric power of the electric refrigerator; The reciprocal of the coefficient of performance of the electric refrigerator; The refrigeration power of the electric refrigerator; The renewable energy consumption constraint condition is: ; Wherein, The photovoltaic predicted power; The actual output of photovoltaic; The photovoltaic power abandoned; The wind power predicted power; The actual power of wind power; The wind power abandoned power; The shared energy storage capacity configuration is: ; Wherein, The maximum configuration capacity of shared energy storage; The capacity configuration coefficient of shared energy storage; The shared energy storage charging power; The gas turbine capacity configuration is: ; Wherein, The maximum configuration power of the gas turbine; The configuration coefficient of the maximum power of the gas turbine; The power generation power of the gas turbine; The grid interaction constraint is: ; ; Wherein, The overall power supply power of the system; is the total electricity purchase power of the data center cluster; is the maximum electricity purchase and sale power; is the total electricity sale power of the system; is the shared energy storage electricity sale power; The system only allows the data center cluster to purchase electricity and the shared energy storage to sell electricity; The global cooling power balance constraint is: ; wherein, is the electricity chiller refrigeration power; is the gas turbine refrigeration power; is the interactive load work cooling coefficient; is the interactive work load electricity consumption power; is the data center processing batch work load cooling power consumption; The global electricity power balance constraint is: ; wherein, is the electricity chiller electricity consumption power; The constraint for guaranteeing the quality of service of the data center is based on the M / M / 1 queuing theory model, which is used to describe the average queuing waiting time of the batch work load; and the non-linear convex function constraint of the M / M / 1 queuing theory model is converted into a group of linear inequality constraints by the tangent approximation method, so as to ensure that the service violation rate of the data center is lower than the preset threshold, and the model is: ; wherein, is the upper limit of the average queuing waiting time of the batch work load; is the acceptable violation rate; is the system utilization rate; task arrival rate; task service rate; is the arrival rate conversion coefficient; is the service rate conversion coefficient; is the original total load; is the adjustable work load; average queuing waiting; average queuing job number.

[0008] Further, the multi-agent collaborative operation model includes a work load model inside the data center, which divides the arrived work load into two categories: interactive work load sensitive to delay and needing to be processed immediately, and batch work load which has flexibility in time scheduling and can be transferred for processing; the batch work load model is:

[0009] wherein, and are the upper and lower bounds of the batch task load scheduling period; is the binary variable of the batch task load processing status; is the electrical power consumed for processing the batch load; is the energy required for processing the j th batch load; is the duration of time consumed for processing the j th batch load; is the refrigeration power generated for processing the batch load; is the electrical to cooling conversion factor for processing the batch load.

[0010] Further, the historical data is processed using a data-driven sample robust optimization method, including: for the constraint of robust upper bound , determine the minimum configuration capacity to meet most of the load demand: ; ; ; for the constraint of robust lower bound , determine the minimum available output to ensure system reliability: ; ; ; wherein, is the historical data observation value, is the binary decision variable used to determine whether to adopt the s th historical sample, is a large enough relaxation parameter; is the preset confidence level; represents the total number of historical observation samples; indicates rounding up, ensuring that at least a proportion of the samples are selected to cover.

[0011] Further, the objective function of the multi-agent collaborative operation model includes: shared energy storage system model: ; ; ; is the shared energy storage arbitrage profit; a shared energy storage degradation cost; a power selling price, a shared energy storage power selling price, a shared energy storage unit degradation cost, a charging power, a discharging power, ; ; ; a data center energy production device operation cost; a photovoltaic data center operation and maintenance cost; a refrigeration device unit operation cost, an electric refrigeration machine power, a gas turbine power, a power purchase power, a power selling price, a, a curtailed wind power cost, a curtailed power, a gas purchase cost.

[0012] a wind power data center:

[0013]

[0014] a wind power data center operation and maintenance cost; a curtailed wind power.

[0015] The constraint condition of the multi-agent collaborative operation is the result after the optimization configuration.

[0016] Further, the distributed optimization algorithm is an alternating direction multiplier method, which decomposes the collaborative operation model into a plurality of local sub-problems corresponding to each data center and the shared energy storage system, and coordinates through a Lagrange multiplier, and iteratively solves under the premise of protecting the internal operation data privacy of each agent.

[0017] Further, the collaborative operation model is decomposed into a plurality of local sub-problems corresponding to each data center and the shared energy storage system, including: establishing a target function of a data center cluster and a shared energy storage multi-agent collaborative operation sub-problem 1: ; wherein, a shared energy storage optimization operation cost, For the first i Optimize operating costs for individual data centers; Constraints of Sub-problem 1: Establishing a multi-entity collaborative operation of data center clusters and shared energy storage: The device capacity constraints obtained from capacity optimization configuration, along with the following formula constraints: ; in, For shared energy storage entities to the first i Expected variables for data center discharge; The expected variable for charging the shared energy storage by the i-th data center entity; The objective function for sub-problem 2, which establishes a multi-entity collaborative operation of a data center cluster and shared energy storage, is as follows: ; in, The independent operating cost of shared energy storage; Sub-problem 1: Shared energy storage cost after optimization; Adjustments for negotiated costs of shared energy storage; Cost of independent operation of the data center; cost of the data center after optimization of sub-problem 1 ; This is an adjustment item for data center negotiation costs; As a bargaining factor; Constraints for Sub-problem 2: Establishing a multi-entity collaborative operation of a data center cluster and shared energy storage: ; .

[0018] Furthermore, a distributed optimization algorithm is used to solve the multi-agent collaborative operation model, including: Constructing the augmented Lagrangian function for subproblem 1 : ; in, For the first i Cost per data center For data centers i Power for charging shared energy storage For shared energy storage to data centers i Discharge power, For the scheduling cycle, For Lagrange multipliers, P1 is the penalty coefficient; shared energy storage and each data center update their own trading power strategy through local calculation, and the entities only exchange trading power information; Initialize primitive variables Initialize dual variables Set the penalty parameter Set the convergence precision threshold Initialize the iteration counter Obtain initial values of the primal variables x and z from the data center cluster and shared energy storage; Let k denote the iteration number; in each iteration, the following steps are performed: Shared energy storage update: ; Data center update: ; Lagrange multiplier update: ; where denotes the Lagrangian function.

[0019] Pass these updated multipliers to all participants for the next iteration; Compute the primal residual and the dual residual and perform a convergence check: ; If convergence is achieved; where is the expected variable of the shared energy storage subject to discharge to the k +1thdata center in the i thiteration; is the expected variable of the k thdata center subject to charge from the shared energy storage in the i +1thiteration; Construct the augmented Lagrangian function of subproblem 2 : ; where is the penalty coefficient of subproblem 2 In each iteration, the following steps are performed: Shared energy storage update: ; Lagrange multiplier update: ; Pass these updated multipliers to all participants for the next iteration; Convergence verification: compute the primal residual and the dual residual and perform a convergence check: ; If converges; where, is the cost saving of the subject i .

[0020] Further, the contribution quantification function is an exponential function based on natural logarithm, which maps the net energy supply of each subject to its bargaining power weight in the game, denoted as: ; ; where, is the bargaining power of each data center; is the total energy provided by each data center in the optimization time domain; is the energy contribution of the data center that makes the maximum contribution; is the total energy obtained by each data center in the optimization time domain; is the energy received by the data center that makes the minimum contribution; is the energy received by the data center that makes the minimum contribution; i to the shared energy storage charging power.

[0021] Advantages and positive effects of the present application: the present application proposes a hierarchical optimization framework for collaborative planning and operation of data center cluster and shared energy storage. The framework couples long-term capacity configuration based on data-driven robust optimization and short-term collaborative scheduling based on cooperative game theory, which not only effectively resists the high uncertainty of renewable energy and realizes the optimization of system full life cycle operation, but also constructs a fair distribution mechanism of cooperative surplus by introducing an asymmetric Nash bargaining model, thereby guaranteeing the long-term stability and fairness of the multi-agent alliance. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0023] Figure 1 is a flow chart of a data center cluster and shared energy storage collaborative operation method in the embodiment of the present application; Figure 2 is a collaborative operation system structure schematic diagram of data center cluster and shared energy storage in the embodiment of the present application; Figure 3Figure 1 is a schematic diagram of a batch load processing process in an embodiment of the present application; Figure 4 Figure 2 is an input data display diagram in an embodiment of the present application; Figure 5 Figure 3 is a configuration process power balance diagram in an embodiment of the present application; Figure 6 Figure 4 is a sub-problem 1 solving iteration diagram in an embodiment of the present application; Figure 7 Figure 5 is a sub-problem 2 solving iteration diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] The present embodiment is composed of a joint system composed of a data center cluster composed of three heterogeneous data centers and a shared energy storage system. Among them, data center 1 and data center 2 are mainly configured with photovoltaic power generation, which are photovoltaic type data centers; data center 3 is mainly configured with wind turbines, which are wind power type data centers. This heterogeneous configuration aims to take advantage of the complementarity of wind and light output. The scheduling period is 24 hours, and the time step is 1 hour.

[0027] As shown in Figure 1 , a data center cluster and shared energy cooperative operation method, specifically comprising the following steps: S1, collect historical data, including hourly photovoltaic output, wind power output and workloads of three data centers.

[0028] Among them, the historical data includes, for example,Figure 3 The data shown covers a whole year, which constitutes a sample space for generating an uncertainty set.

[0029] S2, constructing a data-driven sample robust capacity configuration model; In specific implementation, the following steps are performed: S21, processing historical data by using a data-driven sample robust optimization method; The data-driven sample robust algorithm is specifically: a large number of historical observation data samples are used to construct an uncertainty set that can represent possible future situations. By introducing a confidence level, it is ensured that the decision is feasible under at least the historical scenarios of the confidence level. The confidence level parameter is used to adjust the proportion of samples included in the uncertainty set. By adjusting the confidence level, the decision maker can balance the robustness and economy of the scheme, thereby reducing the conservatism of the decision while ensuring the robustness of the system.

[0030] For the optimization problem, some constraints are affected by uncertain parameters , , s =1, 2, …, n . An uncertainty set is constructed, and S is a sample scenario.

[0031] For the constraint of the robust upper bound , the minimum configuration capacity is determined to meet most of the load demand: ; ; ; For the constraint of the robust lower bound , the minimum available output is determined to ensure system reliability: ; ; ; is a preset confidence level, which is set to 80% in the embodiment, N represents the total number of historical observation samples, represents rounding up, ensuring that at least a proportion of samples are selected to cover. By introducing the robust upper and lower bounds, the model can more comprehensively evaluate and manage uncertainty, ensuring that the system operates under the uncertainty of both computational load and renewable energy.

[0032] S22, a data center and a shared energy storage collaborative configuration objective function is established, and the integrated cost of the system in the entire planning life cycle is minimized, which is converted to the current scheduling period.

[0033] The objective function is as follows: ; is a shared energy storage life cycle conversion factor; is the unit energy capacity investment cost of the shared energy storage; is the maximum energy capacity of the shared energy storage; is a gas turbine life cycle conversion factor; is the unit power capacity investment cost of the gas turbine; is the maximum power capacity of the gas turbine in the data center ; is the purchased power; is the time-of-use electricity price for purchasing electricity; is the sold power; is the time-of-use electricity price for selling electricity; is the energy storage degradation cost coefficient; is the total power of the shared energy storage charging; is the total power of the shared energy storage discharging; is the wind and light curtailment penalty cost coefficient; is the photovoltaic curtailment power; is the wind curtailment power; is the natural gas purchase price; is the gas turbine gas consumption power; is the gas turbine refrigeration cost coefficient; is the uncompleted task penalty coefficient; is the task completion state flag variable.

[0034] S23, the constraint condition of the data center and the shared energy storage collaborative configuration is established; Among them, the gas turbine operation constraint condition is: ; In the formula is the power generation power of the gas turbine; is the maximum power generation power of the gas turbine; is the thermal power of the gas turbine; and the thermal-electricity ratio coefficient of the gas turbine ; The electric refrigerator operation constraint condition is: ; In the formula is the electric power of the electric refrigerator; is the inverse of the coefficient of performance of the electric refrigerator; is the refrigeration power of the electric refrigerator; The renewable energy accommodation constraint condition is: ; In the formula is the photovoltaic predicted power; is the photovoltaic actual output; is the photovoltaic power abandoned; is the wind power predicted power; is the wind power actual power; is the wind power abandoned power; The shared energy storage capacity configuration is: ; In the formula is the maximum configuration capacity of the shared energy storage; is the capacity configuration coefficient of the shared energy storage; is the shared energy storage charging power; The gas turbine capacity configuration is: ; In the formula is the maximum configuration power of the gas turbine; is the configuration coefficient of the maximum power of the gas turbine; is the gas turbine power generation power; The grid interaction constraint is: ; ; is the overall power supply power of the system; is the data center power purchase amount; is the maximum power purchase and sale power; is the overall power sale power of the system; is the shared energy storage power sale power; The system only allows the data center cluster to purchase power and the shared energy storage to sell power; The global cold power balance constraint is: ; is the electric refrigerator refrigeration power; is the gas turbine refrigeration power; is the interactive load power consumption coefficient; is the interactive work load power consumption; is the data center processing batch work load power consumption; The global electric power balance constraint is: ; is the electric refrigerator power consumption.

[0035] The quality of service requirements are as follows: The constraint is based on an M / M / 1 queuing theory model for describing the average queuing waiting time of the batch processing workload; and a set of linear inequality constraints is obtained by converting the nonlinear convex function constraint of the M / M / 1 queuing theory model into a set of linear inequality constraints through a tangent approximation method, so as to ensure that the service violation rate of the data center is lower than a preset threshold, and the model is: ; is an upper limit of the average queuing waiting time of the batch processing workload; is an acceptable violation rate; is a system utilization rate; is a task arrival rate; is a task service rate; is an arrival rate conversion coefficient; is a service rate conversion coefficient; is an original total load; is an adjustable workload; is an average queuing waiting time; is an average queuing job number.

[0036] Two comparative scenarios are set in this embodiment: Scenario 1 (the application): The collaborative configuration method of the application is adopted to simultaneously optimize the capacity of the shared energy storage and the gas turbine.

[0037] Scenario 2 (comparison): No shared energy storage is configured, and only the capacity of the gas turbine inside each data center is optimized.

[0038] Table 1

[0039] The comparison of the system configuration results is shown in Table 1. The total cost of scenario 1 is 19485.01 yuan, while the total cost of scenario 2 is 23400.15 yuan. Although the investment cost (2128.44 yuan) of scenario 1 is higher than that (1277.14 yuan) of scenario 2, the operation cost (17348.24 yuan) of scenario 1 is much lower than that (22123.01 yuan) of scenario 2. This fully proves that the investment in the shared energy storage in the early stage can bring operation cost savings far exceeding the investment through its energy time shifting and flexibility provision in the operation stage, verifying the superiority of the integrated planning and operation idea of the application.

[0040] From the configuration results, the GT configuration capacity (429.76, 277.20, 295.20 kW) of each DC in scenario 1 is significantly lower than that (656.03, 490.38, 305.92 kW) in scenario 2, with a reduction of 35% to 47%. This indicates that the presence of the shared energy storage effectively replaces part of the expensive, fossil fuel-dependent backup capacity.

[0041] S3, establish a data center cluster and shared energy multi-agent collaborative operation model; The goal of the model is to maximize the overall benefits of the alliance, specifically to minimize the total operating cost of the data center cluster and maximize the operating revenue of the shared energy.

[0042] In specific implementation, the following steps are performed: S31, establish a data center cluster and shared energy multi-agent collaborative operation objective function: Shared energy system model: ; ; ; For shared energy arbitrage profit; For shared energy degradation cost.

[0043] Photovoltaic data center model: ; ; ; For data center energy production equipment operating cost; For photovoltaic data center operation and maintenance cost.

[0044] Wind power type data center: ; ; For wind power type data center operation and maintenance cost.

[0045] S32, establish data center cluster and shared energy multi-agent collaborative operation constraint condition: The data center cluster and shared energy multi-agent collaborative operation constraint condition is the constraint condition obtained after the device capacity obtained in the capacity configuration phase is input to the operating model as a constant value.

[0046] S4, according to the asymmetric Nash bargaining theory, the data center cluster and shared energy multi-agent collaborative operation model is decomposed into subproblem 1 and subproblem 2: S41, establish data center cluster and shared energy multi-agent collaborative operation subproblem 1 objective function: ; For shared energy optimization operating cost, Optimize the operation cost for the i-th data center.

[0047] S42, establish the constraint conditions of the data center cluster and shared energy multi-agent collaborative operation sub-problem 1: The constraint conditions are the constraint conditions in S32 and the following formula: ; is the expected variable of the shared energy agent discharging to the i-th data center; is the expected variable of the i-th data center agent charging the shared energy; S43, solve the data center cluster and shared energy multi-agent collaborative operation sub-problem 1 by using ADMM algorithm; The above structure allows the application of ADMM algorithm, which is very effective for solving separable convex optimization problems with coupled linear constraints. ADMM decomposes the global problem into two sub-problems that can be solved locally, and coordinates through the Lagrange multiplier. The augmented Lagrangian function of sub-problem 1 is The following is constructed: ; is the Lagrange multiplier, is the penalty coefficient of P1. The shared energy and each data center update their own trading power strategies through local calculation, and only exchange trading power information between agents. Let k represent the iteration number. In each iteration, the following steps are performed: Shared energy update: ; Data center update: ; Lagrange multiplier update: ; After all the original variables are updated, the Lagrange multiplier is updated according to the violation of the coupling constraint. This update pushes the original variables to a consistent solution that satisfies the coupling constraint, and then these updated multipliers are passed to all participants for the next iteration.

[0048] S44, establish the objective function of the data center cluster and shared energy multi-agent collaborative operation sub-problem 2: ; is the independent operation cost of the shared energy; is the shared energy cost after sub-problem 1 optimization; is the shared energy bargaining cost adjustment term; The data center independent operation cost; the data center cost after sub-problem 1 optimization ; The data center negotiation cost adjustment item The negotiation factor

[0049] After the establishment of the asymmetric Nash negotiation based cooperative game operation model and the optimal cooperative operation strategy is obtained, the cooperation surplus will be generated, that is, the total cost saved compared with all subjects independently running. The goal of this stage is to fairly distribute this surplus.

[0050] Since the natural logarithm is a strictly monotonic increasing convex function, taking the logarithm of the above formula can convert the maximum value problem into a minimum value problem for easy solution. After conversion, the following formula is obtained: ; S45, establish the data center cluster and shared energy storage multi-agent cooperative operation sub-problem 2 constraint condition ; ; S46, solve the data center cluster and shared energy storage multi-agent cooperative operation sub-problem 2 by using ADMM algorithm Similar to sub-problem 1, ADMM algorithm is used to solve sub-problem 2 in a distributed manner. The augmented Lagrangian function of sub-problem 2 is constructed to handle the coupling constraints and promote distributed updates, and the augmented Lagrangian function of sub-problem 2 is established : ; The penalty coefficient of sub-problem 2 is executed in the following steps at each iteration: Shared energy storage update ; Lagrangian multiplier update ; The Lagrangian multiplier is updated according to the violation of the total saving constraint, and the allocated savings are pushed to the total cooperation surplus. Then these updated multipliers are communicated to all participants for the next iteration.

[0051] According to the steps of the data center cluster and shared energy storage cooperative operation method of the application, the model based on the application is used as a benchmark model (scene 3). The model compares the operation of uncoordinated data centers, the model of independent operation of each data center (scene 4) and the operation of uncoordinated data centers, but the cooperative operation of each data center (scene 5). The optimization operation results are shown in Table 2: Table 2

[0052] When sharing energy storage is deployed and cooperation among subjects is considered, the total cost of scenario 3 is the lowest, which is 21520.48 yuan, far lower than 34326.12 yuan of the independent operation mode and 25723.34 yuan of the operation mode without considering sharing energy storage cooperation. This shows that the economic benefit of the system gradually improves from independent operation to cooperative operation, and then to the collaborative operation of introducing sharing energy storage. Among them, sharing energy storage not only helps to reduce the operation cost of each data center, but also creates a profit of 330.54 yuan. Therefore, the conclusion points out that when sharing energy storage is deployed and cooperative operation among subjects is realized, the total cost of the system can be effectively controlled, and the highest reduction of 14.44% can be achieved, which proves the superiority of the collaborative method.

[0053] The collaborative operation method of the data center cluster and the shared energy storage in the above embodiment can effectively overcome the intermittency of renewable energy, build a new low-cost renewable energy supply mode, and effectively solve the problem of cooperative income distribution by adding shared energy storage and asymmetric Nash bargaining model for overall optimization of scheduling strategy.

[0054] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for the coordinated operation of a data center cluster and shared energy storage, characterized in that, To address the operational costs of heterogeneous data center clusters, distributed optimized configuration is achieved using shared energy storage, including the following steps: Collect historical data, including the output of each renewable energy source and the workload of each data center; Based on the historical data, a data-driven robust capacity configuration model is constructed, and the capacity of shared energy storage and gas turbines is optimized using the data-driven robust capacity configuration model; the goal of the robust capacity configuration model is to minimize equipment investment costs and expected operating costs under uncertain scenarios. A multi-entity collaborative operation model is established based on the optimized configuration of the data center cluster and the shared energy storage. The goal of the multi-entity collaborative operation model is to minimize the total operating cost of the data center cluster and maximize the operating benefits of the shared energy storage system. A distributed optimization algorithm is used to solve the multi-entity collaborative operation model to determine the optimal operation strategy of each data center and the shared energy storage system during the scheduling cycle.

2. The method according to claim 1, characterized in that, The multi-entity collaborative operation model is based on an asymmetric Nash bargaining game model, which allocates the cooperative benefits generated during collaborative operation; and calculates the bargaining power of each entity according to the contribution quantification function to determine the final benefit distribution scheme of each entity.

3. The method according to claim 1, characterized in that, Based on the historical data, a data-driven robust capacity configuration model is constructed, including: A data-driven sample robust optimization method is used to process historical data, including: using the historical data to construct an uncertainty set that can represent possible future scenarios; ensuring that the decision is feasible in historical scenarios at least at that confidence level by introducing a confidence level parameter; the confidence level parameter is used to adjust the proportion of samples included in the uncertainty set; Establish a target function for the coordinated configuration of data centers and shared energy storage, wherein the target function is: ； in, To share the energy storage lifecycle conversion factor; Investment cost per unit energy capacity for shared energy storage; The maximum energy capacity for shared energy storage; This is the gas turbine life cycle conversion factor; This refers to the investment cost per unit power capacity of the gas turbine. For data centers i The maximum power capacity of a gas turbine; For the power purchased; Time-of-use pricing for electricity purchases; This refers to the power output for electricity sales. Time-of-use pricing for electricity; This is the energy storage degradation cost coefficient; Total power of shared energy storage charging; To share the total discharge power of the energy storage; The penalty cost coefficient for abandoning scenic views; Solar power curtailment; Wind power curtailment; Natural gas purchase price; Gas turbine power consumption; This refers to the refrigeration cost coefficient for gas turbines. This represents the penalty coefficient for failing to complete the task. This is a variable used to indicate the task completion status. Establish constraints for the coordinated configuration of data centers and shared energy storage, including: The operating constraints of the gas turbine are: ； In the formula This refers to the power generation capacity of the gas turbine. This represents the maximum power output of the gas turbine. The thermal power of the gas turbine; The thermoelectric ratio coefficient of the gas turbine; The operating constraints of the electric chiller are: ； in, For electric refrigeration unit power; It is the reciprocal of the coefficient of performance of the electric refrigeration machine; This refers to the refrigeration power of the electric refrigeration unit. The constraints for renewable energy consumption are: ； in, Forecast power of photovoltaics; To contribute to the actual development of photovoltaics; This refers to the amount of solar power that has been abandoned. Forecast wind power output; This represents the actual power output of the wind power. This refers to the amount of wind power that has been abandoned. The shared energy storage capacity is configured as follows: ； in, Maximum configured capacity for shared energy storage; Capacity configuration factor for shared energy storage; Shared energy storage charging power; The gas turbine capacity configuration is as follows: ； in, This represents the maximum configured power of the gas turbine. The configuration factor is the maximum power output of the gas turbine. For the power generation of gas turbines; The power grid interaction constraints are: ； ； in, Provides the overall power supply for the system; Purchase electricity for data centers; Maximum power consumption for purchasing and selling; This refers to the total electricity sales capacity of the system. To share the power of energy storage for electricity sales; The system only allows data center clusters to purchase electricity and shared energy storage to sell electricity; The global cooling power balance constraint is: ； in, This refers to the refrigeration power of the electric refrigeration unit. This refers to the cooling power of the gas turbine. The operating cooling coefficient is the operating temperature coefficient for interactive loads; Power consumption for interactive workloads; Cooling power consumption for handling batch processing workloads in data centers; The global power balance constraint is: ； in, The power consumption of the electric chiller; The constraints used to ensure the quality of service in the data center are based on the M / M / 1 queuing theory model, which describes the average queuing wait time for batch processing workloads. The nonlinear convex function constraints of the M / M / 1 queuing theory model are transformed into a set of linear inequality constraints using a tangent approximation method to ensure that the service violation rate of the data center is below a preset threshold. The model is as follows: ； in, This represents the upper limit of the average queuing time for batch processing workloads. An acceptable violation rate; For system utilization; Task completion rate; Task service rate; This is the arrival rate conversion factor; Service rate conversion factor; This is the original total load; For adjustable workloads; Average waiting time; Average number of jobs queued.

4. The method according to claim 3, characterized in that, The multi-entity collaborative operation model includes a workload model within the data center. This workload model categorizes incoming workloads into two types: latency-sensitive interactive workloads requiring immediate processing, and batch processing workloads that are flexible in scheduling and can be offloaded. The batch processing workload model is as follows: in, and This serves as the upper and lower bounds for the batch processing task load scheduling period. A binary variable representing the load processing status of batch processing tasks; The electrical power consumed to handle batch processing loads; To process the first j Energy required for a batch processing load; To process the first j The time consumed by each batch processing load; To handle the cooling power generated by batch processing loads; The electrocooling conversion factor is used to handle batch processing loads.

5. The method according to claim 3, characterized in that, A data-driven, sample-based robust optimization method is used to process historical data, including: For robust upper bound Under the constraints, determine the minimum configuration capacity to meet most load demands: ； ； ； For robust lower bound Given the constraints, determine the minimum available output to ensure system reliability: ； ； ； in, These are historical data observations. This is a binary decision variable used to determine whether to adopt the s-th historical sample. A sufficiently large relaxation parameter; It is the preset confidence level; Represents the total number of historical observation samples; This indicates rounding up to the nearest integer, ensuring at least a certain percentage is reached. The selected samples were covered.

6. The method according to claim 1, characterized in that, The objective function of the multi-agent collaborative operation model includes: Shared energy storage system model: ； ； ； To share the profits from energy storage arbitrage; To share the cost of energy storage degradation; For electricity sales price, For the palace maids who sell shared energy storage electricity, To share the unit degradation cost of energy storage, For charging power, For a photovoltaic data center model with high discharge power: ； ； ； Operating costs of energy production equipment in data centers; The operating and maintenance costs of photovoltaic data centers; The unit operating cost of refrigeration equipment, For the power of the electric chiller, For gas turbine power, For the power purchase capacity, For electricity sales price, for, To avoid the cost of abandoning the scenery, For abandoned light power, This refers to the cost of purchasing gas. Wind-powered data centers: For the operation and maintenance costs of wind power-type data centers; This refers to the power of wind curtailment. The constraints for multi-entity collaborative operation are the results of optimized configuration.

7. The method according to claim 1, characterized in that, The distributed optimization algorithm is the alternating direction multiplier method, which decomposes the collaborative operation model into multiple local subproblems corresponding to each data center and the shared energy storage system, and coordinates them through Lagrange multipliers to iteratively solve the problem while protecting the privacy of the internal operating data of each entity.

8. The method according to claim 7, characterized in that, By decomposing the collaborative operation model into multiple local sub-problems corresponding to each data center and the shared energy storage system, including: The objective function for sub-problem 1, which establishes a multi-entity collaborative operation of a data center cluster and shared energy storage, is as follows: ； in, To optimize operating costs for shared energy storage, For the first i Optimize operating costs for individual data centers; Constraints of Sub-problem 1: Establishing a multi-entity collaborative operation of data center clusters and shared energy storage: The device capacity constraints obtained from capacity optimization configuration, along with the following formula constraints: ； in, For shared energy storage entities to the first i Expected variables for data center discharge; The expected variable for charging the shared energy storage by the i-th data center entity; The objective function for sub-problem 2, which establishes a multi-entity collaborative operation of a data center cluster and shared energy storage, is as follows: ； in, The independent operating cost of shared energy storage; Sub-problem 1: Shared energy storage cost after optimization; Adjustments for negotiated costs of shared energy storage; Cost of independent operation of the data center; cost of the data center after optimization of sub-problem 1 ; This is an adjustment item for data center negotiation costs; As a bargaining factor; Constraints for Sub-problem 2: Establishing a multi-entity collaborative operation of a data center cluster and shared energy storage: ； 。 9. The method according to claim 8, characterized in that, The multi-agent collaborative operation model is solved using a distributed optimization algorithm, including: Constructing the augmented Lagrangian function for subproblem 1 : ； in, For the first i Cost per data center For data centers i Power for charging shared energy storage For shared energy storage to data centers i Discharge power, For the scheduling cycle, For Lagrange multipliers, P1 is the penalty coefficient; shared energy storage and each data center update their own trading power strategy through local calculation, and the entities only exchange trading power information; Initialize primitive variables Initialize dual variables Set penalty parameters Set convergence accuracy threshold Initialize the iteration counter It communicates with the data center cluster and shared energy storage to obtain the initial values ​​of the original variables x and z; make k Indicates the number of iterations; in each iteration, the following steps are performed: Shared energy storage upgrade: ； Data Center Update: ； Lagrange multipliers update: ； in, This represents the Lagrange function. These updated multipliers are passed to all participants for the next iteration; Calculate the original residuals and dual residual And perform a convergence check: ； like If so, then convergence occurs; in, For the first k +1st iteration shared energy storage entity to the first i Expected variables for data center discharge; For the first k +1 iterations i Expected variables for each data center entity to charge shared energy storage; Constructing the augmented Lagrangian function for subproblem 2 : ； in, To determine the penalty coefficient for subproblem 2, the following steps are performed in each iteration: Shared energy storage upgrade: ； Lagrange multipliers update: ； These updated multipliers are communicated to all participants for the next iteration; Convergence verification: Calculate the original residuals and dual residual And perform a convergence check: ； like Then it converges; in, as the main body i Cost savings.

10. The method according to claim 2, characterized in that, The contribution quantification function is an exponential function based on the natural logarithm. This function maps the net energy supply of each entity to its bargaining power weight in the game, and is expressed as: ； ； in, This enhances the bargaining power of each data center. The total energy provided by each data center during the optimization time domain; The energy contribution of data centers that make the greatest contribution; The total energy gained by each data center during the optimization time domain; Energy intake of the data center to make the minimum contribution; For data centers i Power to charge shared energy storage.

Images