Multi-data center microgrid scheduling method considering carbon quota and data load interaction

Abstract

The invention relates to the technical field of power dispatching, and provides a multi-data center microgrid dispatching method considering carbon quota and data load interaction in order to solve the problem of multi-data center microgrid distributed and two-stage robust collaborative optimization comprehensively considering carbon quota and data load interaction in an uncertain environment. The method comprises the following steps: step 1, establishing a multi-data center microgrid collaborative operation scene containing carbon quota and data load interaction; step 2, establishing a data center microgrid DCM two-stage robust optimization scheduling model; 3, performing linearization conversion on nonlinear constraints contained in constraint conditions in the scheduling model by adopting a piecewise linearization method; 4, performing equivalent conversion on the ellipsoid uncertainty set by adopting a binary variable expansion method; and 5, on the basis of comprehensively considering the participation of the DCM in carbon quota and data load interaction, performing profit distribution by adopting an asymmetric Nash negotiation model. The method is mainly applied to power dispatching occasions.

Description

Technical Field

[0001] The present invention relates to the technical field of power dispatching, and in particular to a multi-data center microgrid robust collaborative optimization dispatching method considering the interaction between carbon quota and data load. Background Art

[0002] With the iteration and update of new generation information and communication technologies such as 5G communication, edge computing, and big data, the world has entered the era of digital economy. Many provinces and cities have actively carried out and promoted the integration and application of data centers with renewable energy and new energy storage infrastructure to form a new paradigm of data center microgrid (DCM). How to reduce the energy supply cost of data centers through the coordinated optimization of the source-load-storage links of multi-data center microgrids has become a hot issue in current research.

[0003] Unlike traditional microgrids, since the energy consumption of DCM is directly related to the amount of data load it processes, DCM can optimize its own energy consumption by adjusting the interactive processing (IN) load to be processed and the batch processing (BA) load that can be processed later, thereby achieving effective utilization of microgrid resources. However, in the existing research on the collaborative optimization operation of multiple DCMs, most of the literature adopts a centralized optimization method to solve the problem. In practice, different DCMs usually belong to different operating entities, and their internal privacy parameters are difficult to share. Centralized optimization methods are not applicable, and there is a lack of fair and reasonable distribution of the benefits brought by cooperative operation according to the size of contribution.

[0004] In addition, with the establishment and gradual promotion of my country's carbon trading market, carbon quotas have also become an important interactive resource for the coordinated low-carbon optimization operation of multiple DCMs. However, current research on the collaborative optimization of multiple DCMs is mostly focused on coordinating the interaction of pending data loads between different DCMs, but lacks consideration of carbon quota resources, resulting in excessively high DCM carbon trading costs. At the same time, existing studies simply regard the electric-carbon characteristics of each device as linear characteristics, which lacks accuracy. The electric-carbon characteristics of devices such as controllable distributed generation (CDG) based on gas turbines are essentially nonlinear. Therefore, how to deal with the nonlinear electric-carbon characteristics of related equipment in carbon quota interaction is also a problem that needs to be solved.

[0005] In the uncertainty processing of renewable distributed generation (RDG), robust optimization (RO) is often used in the uncertain optimal scheduling of power systems because it does not require detailed probability distribution of uncertain variables or a large number of specific scenarios. However, the essence of RO is to make the best decision under the worst scenario generated by the uncertainty set. Therefore, its economy and robustness depend to a large extent on the construction of the uncertainty set model. The more commonly used model is the box set, in which the prediction errors between different RDGs are assumed to be independent of each other. However, in the same DCM, the prediction errors of different RDGs will be affected by some common environmental factors, which means that there is a certain correlation between the prediction errors of different RDGs. The box set ignores these correlations and contains many scenarios with extremely low probability of occurrence, resulting in excessive conservatism. Another uncertainty set that reflects correlation is the standard ellipsoid set. Although the standard ellipsoid set can cover all historical prediction error scenarios and its conservatism is lower than that of the box set, due to the low probability of occurrence of some prediction error scenarios far away from the center of the ellipsoid, the range of the standard ellipsoid set is still large, and the conservatism needs to be further reduced. Therefore, in order to further balance the robustness and economy of RO results, the uncertainty set model needs to be further improved based on the correlation between the prediction errors of different RDGs. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the present invention aims to solve the problem of distributed and two-stage robust collaborative optimization of multi-data center microgrids that comprehensively consider the interaction between carbon quotas and data loads in an uncertain environment. To this end, the technical solution adopted by the present invention is a multi-data center microgrid scheduling method that considers the interaction between carbon quotas and data loads, and the steps are as follows:

[0007] Step 1: Establish a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction;

[0008] Step 2: Considering the day-ahead constraints and intraday constraints, an improved ellipsoid set with robust adjustment factors for the forecast error of renewable distributed generation RDG and a box uncertainty set for the forecast error of interactive IN load to be processed are constructed to establish a two-stage robust optimization scheduling model for data center microgrid DCM;

[0009] Step 3: Use piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model;

[0010] Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set;

[0011] Step 5: Taking the minimization of the overall scheduling cost of all DCMs as the objective function, while satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the alternating direction multiplier method ADMM algorithm is used for optimization and solution. On the basis of comprehensively considering the interaction between DCM's participation in carbon quotas and data load, the asymmetric Nash negotiation model is used for profit distribution.

[0012] The detailed steps are as follows:

[0013] Step 1: Establish a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction

[0014] Different DCMs are located in different areas. The load of each DCM is the cooling equipment of its servers and computer rooms, which are powered by self-built RDG, CDG, energy storage ESS and the main grid. By utilizing the long-distance transmission characteristics of data loads, different DCMs interact with each other to process data loads, thus realizing cross-regional transfer of energy demand.

[0015] At the same time, under the carbon market mechanism, carbon quotas are shared across regions as virtual resources. Each DCM will be allocated a certain amount of initial carbon quotas. When the initial carbon quota of the DCM is insufficient, it is necessary to purchase additional carbon quotas from the carbon market to ensure that carbon emissions do not exceed the carbon quota.

[0016] Each DCM formulates the power generation and consumption plan and data adjustment plan of each resource in combination with the internal forecast information, and transmits the expected data load to be processed and the carbon quota plan as the boundary information to other DCMs. Each DCM adjusts its own scheduling plan, data load to be processed and carbon quota interaction plan according to the received boundary information, and transmits the adjusted data load to be processed and carbon quota interaction to other DCMs again. The above process is executed cyclically in an iterative manner. After the iteration converges, the interactive balance of the data load to be processed and the carbon quota is achieved. The DCMs calculate the contribution based on the interactive data load to be processed and the carbon quota, and distribute the cooperation benefits according to the contribution.

[0017] Step 2: Considering the day-ahead and intraday constraints, an improved ellipsoid set with robust adjustment factors for RDG forecast error and a box uncertainty set for IN load forecast error are constructed to establish a two-stage robust optimization scheduling model for DCM. The specific steps are as follows:

[0018] Among them, the objective function of the DCM two-stage robust optimization scheduling model is:

[0019]

[0020] and The scheduling cost of DCM i in the day-ahead and day-intraday stages respectively; U iis the uncertainty set, including the uncertainty set of RDG and IN load, where the subscript i is the number item of DCM;

[0021] The objective function of the first stage is to minimize the day-ahead operating cost. Including CDG power generation and backup costs, energy storage charging, discharging and backup costs, power purchase and sales costs from the main grid, and carbon quota interaction costs with the carbon market:

[0022]

[0023] is the contribution of CDG before time t, and are the coefficients of the quadratic and linear terms of CDG power generation cost; and Uplink and downlink backup reserved for CDG, and is the CDG unit uplink and downlink backup cost. and is the charging and discharging power of energy storage, is the unit charge and discharge cost; and Uplink backup reserved for energy storage in charging and discharging states, is the unit upstream reserve cost of energy storage; and It is the downlink reserve reserved for energy storage in charging and discharging states. is the unit downstream backup cost of energy storage, and The power purchased and sold from the main network. and The time-of-use electricity purchase and sale price. and For carbon quotas purchased and sold in the carbon market, c CA ,buy and c CA,sell A unit price for buying and selling carbon allowances;

[0024] The second-stage scheduling problem of DCM is an intraday adjustment problem. The decision variables are the power adjustment of CDG and energy storage, and the unbalanced power that needs to be balanced by the main network. The uncertain variables are the prediction errors of RDG output and data load. The optimization goal of the second stage of DCM is to minimize the unbalanced power penalty of the main network and the data load reduction penalty under the premise of ensuring that the operation constraints are met. For ease of representation, the second-stage decision variables are represented by the prefix "Δ":

[0025]

[0026] and is the positive and negative unbalanced power, and Penalty price for the corresponding unit; is the IN load reduction during the day, ρ cut To reduce the cost per unit data load;

[0027] In order to consider the wind turbine WT fluctuations in the uncertainty set and photovoltaic PV fluctuations The spatial correlation between them is used to reduce the conservatism of the robust optimization results. According to the historical data of RDG prediction errors, an improved ellipsoid set containing robust adjustment factors is constructed:

[0028]

[0029] and are the maximum downward and upward fluctuation errors of WT / PV respectively; the prediction error vector Q i,t is the parameter that determines the shape and size of the ellipsoid set; c i,t is the center of the ellipsoid;

[0030] The conservatism of the two-stage robust optimization results is mainly related to the size of the ellipsoid set. The smaller the set, the lower the conservatism of the results. However, the ellipsoid set also needs to be large enough to contain enough historical data points to ensure robustness. In order to obtain an ellipsoid set with the smallest space and that can contain enough historical prediction error scenarios, the following optimization problem is constructed to solve Q i,t and c i,t :

[0031]

[0032] Where: The hth prediction error vector and is the h-th historical forecast error data of WT and PV; Γ is the robust adjustment factor between (0,1];

[0033] The prediction accuracy of IN load is usually high, and its prediction error Using box-type uncertain set representation:

[0034]

[0035] and The maximum upward and downward fluctuation range of IN load;

[0036] The constraints of the two-stage robust optimization scheduling model of DCM include day-ahead constraints and intraday constraints;

[0037] (1) Day-ahead constraints include data load interaction and regulation constraints, energy consumption equipment constraints, energy supply equipment constraints, and carbon emission constraints:

[0038] 1) Data load interaction and adjustment constraints

[0039] When DCMs exchange pending loads, it is necessary to ensure that the pending IN / BA load of each DCM after interaction is greater than 0, and the received pending load does not exceed the maximum receiving bandwidth of DCM i:

[0040]

[0041]

[0042] and is the pending IN / BA load of DCM i before and after the interactive pending data load, is the pending IN / BA load exchanged between DCM i and DCM j. If it is greater than 0, it means that DCM i transfers the pending IN / BA load to DCM j. D is the size of a single data load. i,max is the maximum receiving bandwidth of DCM i;

[0043] IN load needs to be processed in time, and the actual IN load processed in each period satisfy:

[0044]

[0045] BA load can be processed with a delay, but must be within the maximum allowed delay The processing is completed within 10 seconds, so the actual processing BA load The following constraints are met:

[0046]

[0047] To ensure the quality of data load processing, according to the data load queue model and service level agreement requirements, the data load processed in each time period must also meet the queuing delay constraints:

[0048]

[0049] is the number of servers turned on. Since the number of servers in DCM is usually large, this variable is treated as a continuous variable. is the maximum number of servers allowed to be opened in DCM i; μ i is the processing rate of a single server; is the maximum allowed queuing delay;

[0050] 2) Energy-consuming equipment constraints

[0051] Energy-consuming equipment in DCM includes servers and refrigeration air conditioners, among which servers consume Mainly includes processing energy consumption and receiving energy consumption Both energy consumptions are related to the number of servers powered on and the load they receive to be processed:

[0052]

[0053] and P i peak They are the idle and peak power of a single server respectively; and They correspond to the fixed energy consumption of a single server receiving device and the receiving energy consumption per unit of data load to be processed;

[0054] To ensure that the temperature in the equipment room is within the allowable range, the equipment room cooling and air conditioning should meet the following requirements:

[0055]

[0056] and are the internal and external temperatures of the room, and The minimum and maximum permissible room internal temperatures are h and c h is the equivalent thermal resistance and thermal capacity of the computer room; is the heat generation coefficient of the server, The energy efficiency coefficient of variable frequency air conditioning refrigeration; is the energy consumption of variable frequency air conditioner; is the maximum energy consumption of the inverter air conditioner;

[0057] 3) Energy supply equipment constraints

[0058] DCM's energy supply equipment includes RDG, CDG and energy storage. Except for RDG, all other equipment are controllable. Since CDG power generation is adjustable, CDG needs to provide a certain amount of backup capacity for the intraday stage in addition to power generation to cope with fluctuations in RDG and data center power loads:

[0059]

[0060] and is the minimum and maximum output of CDG, is the maximum climbing rate of CDG;

[0061] Since the charging and discharging power of energy storage is adjustable, in addition to charging and discharging, energy storage can also reserve backup capacity for the intraday phase in the day-ahead phase, just like CDG. When considering energy storage to provide backup, the operating constraints of energy storage in the day-ahead phase are as follows:

[0062]

[0063] The maximum charging and discharging power of energy storage; and A binary variable representing the energy storage state. When it is 1, it means that the energy storage is in the charging and discharging state respectively. M is a sufficiently large constant.

[0064] The state of charge constraint of energy storage can be expressed as:

[0065]

[0066] It is the energy storage charging and discharging efficiency; and The lower and upper limits of the energy storage charge state;

[0067] When the power generation of CDG, RDG and energy storage in DCM is not equal to the load energy consumption, DCM can also ensure the day-ahead power balance by exchanging power with the upper main grid:

[0068]

[0069] and Provides day-ahead forecasting for WT and PV within DCM;

[0070] 4) Carbon emission constraints

[0071] DCM carbon emissions mainly come from purchasing electricity from the main grid and CDG power generation:

[0072]

[0073] is the total carbon emission of DCM i during the dispatch period, and are the carbon emissions from purchasing electricity from the main grid and generating electricity from CDG, where the carbon emissions from CDG are nonlinear functions. and is the secondary and primary carbon emission coefficient of CDG; b buy,E The primary carbon emission coefficient for purchasing electricity from the main grid;

[0074] The carbon emission constraint of DCM in the day-ahead phase is expressed as:

[0075]

[0076] is the initial carbon quota of DCM i; is the carbon quota exchanged between DCM i and DCM j. If it is greater than 0, it means that DCMi shares the carbon quota with DCM j. The subscripts i and j are the numbers of DCMs, which are used to distinguish different DCMs. CA is the carbon quota margin coefficient; and are the carbon allowances purchased and sold by DCM i, respectively;

[0077] (2) Intraday constraints include load processing and equipment energy consumption constraints, energy supply equipment regulation and power balance constraints:

[0078] 1) Load processing and equipment energy constraints

[0079] The fluctuation of data load mainly comes from the prediction error of IN load, and the IN load processing adjustment amount The following constraints are met:

[0080]

[0081] Data load regulation will also cause fluctuations in energy consumption of various devices in the data center, and server energy consumption fluctuations The calculation method is as follows:

[0082]

[0083] and They are the energy consumption fluctuations of the data replication processing and receiving equipment, respectively;

[0084] In order to ensure the stability of the room temperature, the amount of adjustment of the air conditioning cooling power during the day needs to offset the fluctuation of server heating:

[0085]

[0086] It is the adjustment amount of refrigeration and air conditioning power consumption during the day;

[0087] 2) Energy supply equipment regulation and power balance constraints

[0088] CDG and energy storage can adjust the output power within the reserve capacity range. The intraday power regulation constraints of CDG and energy storage are as follows:

[0089]

[0090] and is the uplink and downlink adjustment amount of CDG power; and It is the uplink and downlink regulation amount of charge / discharge power;

[0091] In addition, power balance constraints must be met during the intraday phase:

[0092]

[0093] Step 3: Use piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model;

[0094] The nonlinear constraints contained in the constraints of the scheduling model are mainly the carbon emission constraints of each DCM, which are linearized by piecewise linearization and expressed as:

[0095]

[0096] For about The convex function of , so the above formula is further relaxed to:

[0097]

[0098] exist Take S+1 points on the function curve with equal horizontal coordinates, and mark the coordinates as (p s ,w s ), where s = 0, 1, 2, ..., S. Connect all adjacent points with straight lines, then The curve is approximated by S line segments; therefore, the carbon emission constraints of each DCM are further transformed into S linear inequality constraints, where the sth constraint is:

[0099]

[0100] s=0,1,2,…,S;

[0101] Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set. The detailed steps are as follows:

[0102] Using uncertain variables represents the total RDG prediction error, which is the sum of WT and PV prediction errors, The uncertain set is expressed as:

[0103]

[0104] and As auxiliary variables representing the fluctuation range of RDG prediction error, since the RDG output fluctuation is the largest in the worst scenario, the above two variables can be processed as binary variables; and is the maximum upward and downward fluctuation range of the RDG prediction error, after considering the correlation between WT and PV prediction errors in step 2, and Use the CPLEX solver to solve the problem according to the following formula:

[0105]

[0106]

[0107] Introducing binary variables and After that, the equivalent box uncertainty set of IN load can be expressed as:

[0108]

[0109] Step 5: Taking the minimum overall dispatch cost of all DCMs as the objective function, while satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the ADMM algorithm is used for optimization and solution. On the basis of comprehensively considering the interaction between the data load to be processed and the carbon quota of DCMs, the asymmetric Nash negotiation model is used for profit distribution. The detailed steps are as follows:

[0110] The objective function is to minimize the overall scheduling cost of all DCMs, while satisfying the constraints of the data load to be processed and the interaction consistency of carbon quotas between DCMs:

[0111]

[0112] x i is the variable for independent decision making of DCM i, v i,j is the coupling variable between DCM i and DCM j, specifically the data load to be processed for interaction between DCMs and interactive carbon quotas N DCM is the DCM set;

[0113] In order to ensure the privacy of each subject, the ADMM method is used to solve the collaborative optimization scheduling problem of multiple DCMs. The coupling variables v are exchanged between each DCM. i,j The optimization result is used to update the local sub-problem model and continuously iterate the solution; when the coupling variables exchanged by adjacent DCMs are close enough, the iteration is completed and the best scheduling solution can be obtained. In the kth ADMM iteration, the optimization scheduling problem of the relaxed DCM i is expressed as:

[0114]

[0115] ρ is the quadratic penalty factor of the augmented Lagrangian function, is the average value of the coupled variable in the kth iteration, which can be updated as follows after k iterations:

[0116]

[0117] is the boundary variable v i,j The optimization result after the kth iteration, λ i,k is the Lagrange multiplier. To ensure the convergence of ADMM, λ i,k After each iteration, it is updated as follows:

[0118]

[0119] When the coupling constraints are satisfied within a certain error and the obtained results are stable enough, the original residual δ P and the dual residual δ D When both are less than the given value, the ADMM iteration process stops, δ P and δ D Calculate according to the following formula:

[0120]

[0121] The comprehensive contribution is calculated based on the data load to be processed and the amount of carbon quotas interacted by each DCM, and the profit is distributed using the asymmetric Nash negotiation model. The contribution of each DCM is r i The definition is as follows:

[0122]

[0123] γ DL and γ CA are the weights of the interaction between the data load to be processed and the carbon quota interaction, satisfying γ DL +γ CA =1;

[0124] definition and are the operating costs of DCM i before and after it participates in the interaction between the data load to be processed and the carbon quota, For the benefits allocated to them through interaction, the profit distribution model based on the asymmetric Nash bargaining model is expressed as follows:

[0125]

[0126] The characteristics and beneficial effects of the present invention are:

[0127] The invention proposes a multi-DCM distributed and two-stage robust optimization model that considers the interaction between carbon quotas and data loads to balance the robustness and economy of the two-stage robust optimization results and achieve a fair and reasonable distribution of cooperation benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] Figure 1 Establish a framework diagram of the collaborative operation scenario of multi-data center microgrids with carbon quota and data load interaction. DETAILED DESCRIPTION

[0129] In view of the shortcomings of the existing technology, the present invention proposes a multi-DCM distributed and two-stage robust optimization model that considers the interaction between carbon quota and data load, so as to balance the robustness and economy of the two-stage robust optimization results and achieve fair and reasonable distribution of cooperation benefits.

[0130] The present invention provides a multi-data center microgrid robust collaborative optimization scheduling method considering the interaction between carbon quota and data load, and the method comprises the following steps.

[0131] Step 1: Establish a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction.

[0132] Step 2: Considering the day-ahead and intraday constraints, an improved ellipsoid set with robust adjustment factors for RDG forecast error and a box uncertainty set for IN load forecast error are constructed to establish a two-stage robust optimization scheduling model for DCM.

[0133] Step 3: Use the piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model.

[0134] Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set.

[0135] Step 5: Taking the minimization of the overall scheduling cost of all DCMs as the objective function and satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the alternating direction method of multipliers (ADMM) algorithm is used for optimization and solution. On the basis of comprehensively considering the interaction between DCMs’ participation in carbon quotas and data loads, the asymmetric Nash negotiation model is used for profit distribution.

[0136] The present invention is further described in detail below with reference to the accompanying drawings and specific examples.

[0137] Step 1: If Figure 1 As shown, a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction is established.

[0138] Different DCMs are located in different areas. The load of each DCM is the cooling equipment of its servers and computer rooms, which are powered by self-built RDG (photovoltaic (PV) and wind turbine (WT)), CDG, energy storage ESS and the main grid. Taking advantage of the long-distance transmission of data loads, different DCMs can interact with each other to process data loads, thereby realizing cross-regional transfer of energy demand.

[0139] At the same time, under the carbon market mechanism, carbon quotas can also be shared across regions as virtual resources. Each DCM will be allocated a certain amount of initial carbon quotas. When the initial carbon quota of the DCM is insufficient, it is necessary to purchase additional carbon quotas from the carbon market to ensure that carbon emissions do not exceed the carbon quota.

[0140] Each DCM formulates the power generation and consumption plan and data adjustment plan for each resource based on the internal forecast information, and transmits the expected data load to be processed and carbon quota plan as boundary information to other DCMs. Each DCM adjusts its own scheduling plan, data load to be processed, and carbon quota interaction plan based on the received boundary information, and retransmits the adjusted data load to be processed and carbon quota interaction to other DCMs. The above process is executed cyclically in an iterative manner. After the iteration converges, the interactive balance of the data load to be processed and the carbon quota is achieved. The DCMs calculate the contribution based on the interactive data load to be processed and the carbon quota, and distribute the cooperation benefits according to the contribution.

[0141] Step 2: Considering the day-ahead and intraday constraints, an improved ellipsoid set with robust adjustment factors for RDG forecast error and a box uncertainty set for IN load forecast error are constructed to establish a two-stage robust optimization scheduling model for DCM.

[0142] Among them, the objective function of the DCM two-stage robust optimization scheduling model is:

[0143]

[0144] and The scheduling cost of DCM i in the day-ahead and day-intraday stages respectively; U i is the uncertainty set, including the uncertainty set of RDG and IN load.

[0145] The objective function of the first stage is to minimize the day-ahead operating cost. Including CDG power generation and backup costs, energy storage charging, discharging and backup costs, power purchase and sales costs from the main grid, and carbon quota interaction costs with the carbon market:

[0146]

[0147] Thank you for your contribution to CDG's recent achievements. and are the coefficients of the quadratic and linear terms of CDG power generation cost; and Uplink and downlink backup reserved for CDG, and is the CDG unit uplink and downlink backup cost. and is the charging and discharging power of energy storage, is the unit charge and discharge cost; and Uplink backup reserved for energy storage in charging and discharging states, is the unit upstream reserve cost of energy storage; and It is the downlink reserve reserved for energy storage in charging and discharging states. is the unit downstream backup cost of energy storage. and The power purchased and sold from the main network. and The price for purchasing and selling electricity based on time of use. and For carbon quotas purchased and sold in the carbon market, c CA,buy and c CA,sell The unit price for buying and selling carbon allowances.

[0148] The second-stage scheduling problem of DCM is an intraday adjustment problem. The decision variables are the power adjustment of CDG and energy storage, and the unbalanced power that needs to be balanced by the main network. The uncertain variables are the prediction errors of RDG output and data load. The optimization goal of the second stage of DCM is to minimize the unbalanced power penalty of the main network and the data load reduction penalty in this scenario while ensuring that the operating constraints are met. For ease of representation, the second-stage decision variables are represented by the prefix "Δ".

[0149]

[0150] and is the positive and negative unbalanced power, and Penalty price for the corresponding unit; is the IN load reduction during the day, ρ cut To reduce the cost per unit data load.

[0151] In order to consider WT fluctuations in the uncertainty set and PV fluctuations The spatial correlation between them is used to reduce the conservatism of the robust optimization results. According to the historical data of RDG prediction errors, an improved ellipsoid set containing robust adjustment factors is constructed:

[0152]

[0153] and are the maximum downward and upward fluctuation errors of WT / PV respectively; the prediction error vector Q i,t is the parameter that determines the shape and size of the ellipsoid set; c i,t is the center of the ellipsoid.

[0154] The conservatism of the two-stage robust optimization results is mainly related to the size of the ellipsoid set. The smaller the set, the less conservative the result. However, the ellipsoid set also needs to be large enough to contain enough historical data points to ensure robustness. In order to obtain an ellipsoid set with the smallest space and that can contain enough historical prediction error scenarios, the optimization problem Q is constructed as shown in the following formula: i,t and c i,t :

[0155]

[0156] Where: The hth prediction error vector and is the hth historical forecast error data of WT and PV. Γ is a robust adjustment factor between (0,1].

[0157] The prediction accuracy of IN load is usually high, and its prediction error It can be represented by a box-type uncertain set:

[0158]

[0159] and It is the maximum upward and downward fluctuation range of IN load.

[0160] The constraints of the two-stage DCM robust optimization scheduling model include day-ahead constraints and intraday constraints.

[0161] (1) Day-ahead constraints include data load interaction and regulation constraints, energy consumption equipment constraints, energy supply equipment constraints, and carbon emission constraints:

[0162] 1) Data load interaction and adjustment constraints

[0163] When DCMs exchange pending loads, it is necessary to ensure that the pending IN / BA load of each DCM after interaction is greater than 0, and the received pending load does not exceed the maximum receiving bandwidth of DCM i:

[0164]

[0165]

[0166] and It is the IN / BA load to be processed of DCM i before and after the interactive data load to be processed. is the pending IN / BA load exchanged between DCM i and DCM j. If it is greater than 0, it means that DCM i transfers the pending IN / BA load to DCM j. D is the size of a single data load; B i,max is the maximum receiving bandwidth of DCM i.

[0167] IN load needs to be processed in time, and the actual IN load processed in each period The following constraints are met:

[0168]

[0169] BA load can be processed with a delay, but must be within the maximum allowed delay Therefore, the actual BA load satisfy:

[0170]

[0171] To ensure the quality of data load processing, according to the data load queue model and service level agreement requirements, the data load processed in each time period must also meet the queuing delay constraints:

[0172]

[0173] is the number of servers turned on. Since the number of servers in DCM is usually large, this variable can be treated as a continuous variable. is the maximum number of servers allowed to be opened in DCM i; μ i is the processing rate of a single server; is the maximum allowed queuing delay;

[0174] 2) Energy-consuming equipment constraints

[0175] Energy-consuming equipment in DCM includes servers and refrigeration and air conditioning. Mainly includes processing energy consumption and receiving energy consumption Both energy consumptions are related to the number of servers powered on and the load they receive to be processed:

[0176]

[0177] and They are the idle and peak power of a single server respectively; and They correspond to the fixed energy consumption of a single server receiving device and the receiving energy consumption per unit of data load to be processed.

[0178] To ensure that the temperature in the equipment room is within the allowable range, the equipment room cooling and air conditioning should meet the following requirements:

[0179]

[0180] and are the internal and external temperatures of the room, and The minimum and maximum permissible room internal temperatures are h and c h is the equivalent thermal resistance and thermal capacity of the computer room; is the heat generation coefficient of the server, The energy efficiency coefficient of variable frequency air conditioning refrigeration; is the energy consumption of variable frequency air conditioner; is the maximum energy consumption of the inverter air conditioner;

[0181] 3) Energy supply equipment constraints

[0182] DCM's energy supply equipment includes RDG, CDG and energy storage, and all except RDG are controllable equipment. Since CDG power generation is adjustable, CDG needs to provide a certain amount of backup capacity for the intraday stage in addition to power generation to cope with fluctuations in RDG and data center power loads:

[0183]

[0184] and is the minimum and maximum output of CDG, is the maximum climbing rate of CDG.

[0185] Since the charging and discharging power of energy storage is adjustable, in addition to charging and discharging, energy storage can also reserve backup capacity for the intraday phase in the day-ahead phase, just like CDG. When considering energy storage to provide backup, the operating constraints of energy storage in the day-ahead phase are as follows:

[0186]

[0187] The maximum charging and discharging power of energy storage; and A binary variable representing the state of energy storage. When it is 1, it means that the energy storage is in the charging and discharging state respectively; M is a sufficiently large constant.

[0188] The state of charge constraint of energy storage can be expressed as:

[0189]

[0190] It is the energy storage charging and discharging efficiency; and It is the lower and upper limit of the energy storage charge state.

[0191] When the power generation of CDG, RDG and energy storage in DCM is not equal to the load energy consumption, DCM can also ensure the day-ahead power balance by exchanging power with the upper main grid:

[0192]

[0193] and Provides day-ahead forecasting for WT and PV in DCM.

[0194] 4) Carbon emission constraints

[0195] DCM carbon emissions mainly come from purchasing electricity from the main grid and CDG power generation:

[0196]

[0197] is the total carbon emission of DCM i during the scheduling period. and They are the carbon emissions from purchasing electricity from the main grid and generating electricity from CDG, where the carbon emissions from CDG are a nonlinear function. and is the secondary and primary carbon emission coefficient of CDG; b buy,E The primary carbon emission coefficient for purchasing electricity from the main grid.

[0198] The carbon emission constraint of DCM in the day-ahead phase can be expressed as:

[0199]

[0200] is the initial carbon quota of DCM i; is the carbon quota exchanged between DCM i and DCM j. If it is greater than 0, it means that DCMi shares the carbon quota with DCM j. CA is the carbon quota margin coefficient; and are the carbon allowances purchased and sold by DCM i, respectively;

[0201] (2) Intraday constraints include load processing and equipment energy consumption constraints, energy supply equipment regulation and power balance constraints:

[0202] 1) Load processing and equipment energy constraints

[0203] The fluctuation of data load mainly comes from the prediction error of IN load, and the IN load processing adjustment amount The following constraints are met:

[0204]

[0205] Data load regulation will also cause fluctuations in energy consumption of various devices in the data center, and server energy consumption fluctuations The calculation method is as follows:

[0206]

[0207] and They are the energy consumption fluctuations of data replication processing and receiving equipment, respectively.

[0208] In order to ensure the stability of the room temperature, the present invention considers that the adjustment amount of the air conditioning cooling power during the day stage needs to offset the fluctuation of the server heating:

[0209]

[0210] It is the adjustment amount of refrigeration and air conditioning power consumption during the day.

[0211] 2) Energy supply equipment regulation and power balance constraints

[0212] CDG and energy storage can adjust the output power within the reserve capacity range. The intraday power regulation constraints of CDG and energy storage are as follows:

[0213]

[0214] and is the uplink and downlink adjustment amount of CDG power; and It is the upstream and downstream adjustment amount of charge / discharge power.

[0215] In addition, power balance constraints must be met during the intraday phase:

[0216]

[0217] Step 3: Use the piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model.

[0218] The nonlinear constraints contained in the constraints of the scheduling model are mainly the carbon emission constraints of each DCM, which are linearized by piecewise linearization and expressed as:

[0219]

[0220] For about is a convex function. Therefore, the above formula can be further relaxed as:

[0221]

[0222] exist Take S+1 points on the function curve with equal horizontal coordinates, and mark the coordinates as (p s ,w s ), where s = 0, 1, 2, ..., S. Connect all adjacent points with straight lines, then The curve can be approximated by S line segments. Therefore, the carbon emission constraints of each DCM can be further transformed into S linear inequality constraints, where the sth (s = 0, 1, 2, ..., S) constraint is:

[0223]

[0224] Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set.

[0225] Using uncertain variables represents the total RDG prediction error, which is the sum of WT and PV prediction errors, The uncertain set can be expressed as:

[0226]

[0227] and As auxiliary variables representing the fluctuation range of RDG prediction error, since the RDG output fluctuation is the largest in the worst scenario, the above two variables can be processed as binary variables; and is the maximum upward and downward fluctuation range of the RDG prediction error. After considering the correlation between WT and PV prediction errors in step 2, and The CPLEX solver can be used to solve the problem as follows:

[0228]

[0229]

[0230] Introducing binary variables and After that, the equivalent box uncertainty set of IN load can be expressed as:

[0231]

[0232] Step 5: Taking the minimization of the overall scheduling cost of all DCMs as the objective function and satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the ADMM algorithm is used for optimization and solution. On the basis of comprehensively considering the DCM participation in the data load to be processed and the carbon quota interaction, the asymmetric Nash negotiation model is used for profit distribution.

[0233] The objective function is to minimize the overall scheduling cost of all DCMs, while satisfying the constraints of the data load to be processed and the interaction consistency of carbon quotas between DCMs:

[0234]

[0235] x i is the independent decision variable of DCM i, X i is the set of independent decision variables of DCM i; v i,j is the coupling variable between DCM i and DCM j, specifically the data load to be processed for interaction between DCMs and interactive carbon quotas V i,j is the set of coupled variables; N DCM It is a DCM collection.

[0236] In order to ensure the privacy of each subject, the ADMM method is used to solve the collaborative optimization scheduling problem of multiple DCMs. The coupling variables v are exchanged between each DCM. i,j The optimization result is used to update the local sub-problem model and continuously iterate to solve it. When the coupling variables exchanged by adjacent DCMs are close enough, the iteration is completed and the best scheduling solution can be obtained. In the kth ADMM iteration, the optimization scheduling problem of the relaxed DCM i can be expressed as:

[0237]

[0238] ρ is the quadratic penalty factor of the augmented Lagrangian function, is the average value of the coupled variable in the kth iteration, which can be updated as follows after k iterations:

[0239]

[0240] is the boundary variable v i,j The optimization result after the kth iteration. i,k is the Lagrange multiplier. To ensure the convergence of ADMM, λ i,kAfter each iteration, it is updated as follows:

[0241]

[0242] When the coupling constraints are satisfied within a certain error and the obtained results are stable enough, the original residual δ P and the dual residual δ D When both are less than the given value, the ADMM iteration process stops. P and δ D Calculate according to the following formula:

[0243]

[0244] The comprehensive contribution is calculated based on the data load to be processed and the amount of carbon quotas interacted by each DCM, and the profit is distributed using the asymmetric Nash negotiation model. i The definition is as follows:

[0245]

[0246]

[0247] γ DL and γ CA are the weights of the interaction between the data load to be processed and the carbon quota interaction, satisfying γ DL +γ CA =1.

[0248] definition and are the operating costs of DCM i before and after it participates in the interaction between the data load to be processed and the carbon quota, The profit distribution model based on the asymmetric Nash bargaining model is as follows:

[0249]

[0250] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-data center microgrid scheduling method considering the interaction between carbon quota and data load, characterized in that the steps are: as follows: Step 1: Establish a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction; Step 2: Considering the day-ahead constraints and intraday constraints, an improved ellipsoid set with robust adjustment factors for the forecast error of renewable distributed generation RDG and a box uncertainty set for the forecast error of interactive IN load to be processed are constructed to establish a two-stage robust optimization scheduling model for data center microgrid DCM; Step 3: Use piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model; Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set; Step 5: Taking the minimization of the overall scheduling cost of all DCMs as the objective function, while satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the alternating direction multiplier method ADMM algorithm is used for optimization and solution. On the basis of comprehensively considering the interaction between DCM's participation in carbon quotas and data load, the asymmetric Nash negotiation model is used for profit distribution.

2. The multi-data center microgrid scheduling method considering the interaction between carbon quota and data load as claimed in claim 1 is characterized in that: The detailed steps are as follows: Step 1: Establish a multi-data center microgrid collaborative operation scenario with carbon quota and data load interaction Different DCMs are located in different areas. The load of each DCM is the cooling equipment of its servers and computer rooms, which are powered by self-built RDG, CDG, energy storage ESS and the main grid. By utilizing the long-distance transmission characteristics of data loads, different DCMs interact with each other to process data loads, thus realizing cross-regional transfer of energy demand. At the same time, under the carbon market mechanism, carbon quotas are shared across regions as virtual resources. Each DCM will be allocated a certain amount of initial carbon quotas. When the initial carbon quota of the DCM is insufficient, it is necessary to purchase additional carbon quotas from the carbon market to ensure that carbon emissions do not exceed the carbon quota. Each DCM formulates the power generation and consumption plan and data adjustment plan of each resource in combination with the internal forecast information, and transmits the expected data load to be processed and the carbon quota plan as the boundary information to other DCMs. Each DCM adjusts its own scheduling plan, data load to be processed and carbon quota interaction plan according to the received boundary information, and transmits the adjusted data load to be processed and carbon quota interaction to other DCMs again. The above process is executed cyclically in an iterative manner. After the iteration converges, the interactive balance of the data load to be processed and the carbon quota is achieved. The DCMs calculate the contribution based on the interactive data load to be processed and the carbon quota, and distribute the cooperation benefits according to the contribution. Step 2: Considering the day-ahead and intraday constraints, an improved ellipsoid set with robust adjustment factors for RDG forecast error and a box uncertainty set for IN load forecast error are constructed to establish a two-stage robust optimization scheduling model for DCM. The specific steps are as follows: Among them, the objective function of the DCM two-stage robust optimization scheduling model is: and The scheduling cost of DCM i in the day-ahead and day-intraday stages respectively; U i is the uncertainty set, including the uncertainty set of RDG and IN load, where the subscript i is the number item of DCM; The objective function of the first stage is to minimize the day-ahead operating cost. Including CDG power generation and backup costs, energy storage charging, discharging and backup costs, power purchase and sales costs from the main grid, and carbon quota interaction costs with the carbon market: is the contribution of CDG before time t, and are the coefficients of the quadratic and linear terms of CDG power generation cost; and Uplink and downlink backup reserved for CDG, and is the CDG unit uplink and downlink backup cost. and is the charging and discharging power of energy storage, is the unit charge and discharge cost; and Uplink backup reserved for energy storage in charging and discharging states, is the unit upstream reserve cost of energy storage; and It is the downlink reserve reserved for energy storage in charging and discharging states. is the unit downstream backup cost of energy storage, and The power purchased and sold from the main network. and is the time-of-use electricity purchase and sale price, W i buy and W i sell For carbon quotas purchased and sold in the carbon market, c CA,buy and c CA,sell A unit price for buying and selling carbon allowances; The second-stage scheduling problem of DCM is an intraday adjustment problem. The decision variables are the power adjustment of CDG and energy storage, and the unbalanced power that needs to be balanced by the main network. The uncertain variables are the prediction errors of RDG output and data load. The optimization goal of the second stage of DCM is to minimize the unbalanced power penalty of the main network and the data load reduction penalty under the premise of ensuring that the operation constraints are met. For ease of representation, the second-stage decision variables are represented by the prefix "Δ": and is the positive and negative unbalanced power, and Penalty price for the corresponding unit; is the IN load reduction during the day, ρ cut To reduce the cost per unit data load; In order to consider the wind turbine WT fluctuations in the uncertainty set and photovoltaic PV fluctuations The spatial correlation between them is used to reduce the conservatism of the robust optimization results. According to the historical data of RDG prediction errors, an improved ellipsoid set containing robust adjustment factors is constructed: and are the maximum downward and upward fluctuation errors of WT / PV respectively; the prediction error vector Q i,t is the parameter that determines the shape and size of the ellipsoid set; c i,t is the center of the ellipsoid; The conservatism of the two-stage robust optimization results is mainly related to the size of the ellipsoid set. The smaller the set, the lower the conservatism of the results. However, the ellipsoid set also needs to be large enough to contain enough historical data points to ensure robustness. In order to obtain an ellipsoid set with the smallest space and that can contain enough historical prediction error scenarios, the following optimization problem is constructed to solve Q i,t and c i,t : Where: The hth prediction error vector and is the h-th historical forecast error data of WT and PV; Γ is the robust adjustment factor between (0,1]; The prediction accuracy of IN load is usually high, and its prediction error Using box-type uncertain sets: and The maximum upward and downward fluctuation range of IN load; The constraints of the two-stage robust optimization scheduling model of DCM include day-ahead constraints and intraday constraints; (1) Day-ahead constraints include data load interaction and regulation constraints, energy consumption equipment constraints, energy supply equipment constraints, and carbon emission constraints: 1) Data load interaction and adjustment constraints When DCMs exchange pending loads, it is necessary to ensure that the pending IN / BA load of each DCM after interaction is greater than 0, and the received pending load does not exceed the maximum receiving bandwidth of DCM i: and is the pending IN / BA load of DCM i before and after the interactive pending data load, is the pending IN / BA load exchanged between DCM i and DCM j. If it is greater than 0, it means that DCM i transfers the pending IN / BA load to DCM j. D is the size of a single data load. i,max is the maximum receiving bandwidth of DCM i; IN load needs to be processed in time, and the actual IN load processed in each period satisfy: BA load can be processed with a delay, but must be within the maximum allowed delay The processing is completed within 1 second, so the actual processing BA load The following constraints are met: To ensure the quality of data load processing, according to the data load queue model and service level agreement requirements, the data load processed in each time period must also meet the queuing delay constraints: is the number of servers turned on. Since the number of servers in DCM is usually large, this variable is treated as a continuous variable. is the maximum number of servers allowed to be opened in DCM i; μ i is the processing rate of a single server; is the maximum allowed queuing delay; 2) Energy-consuming equipment constraints Energy-consuming equipment in DCM includes servers and refrigeration air conditioners, among which servers consume Mainly includes processing energy consumption and receiving energy consumption Both energy consumptions are related to the number of servers powered on and the load they receive to be processed: P i idle and P i peak are the idle and peak power of a single server respectively; P i re,fix and P i re,net They correspond to the fixed energy consumption of a single server receiving device and the receiving energy consumption per unit of data load to be processed; To ensure that the temperature in the equipment room is within the allowable range, the equipment room cooling and air conditioning should meet the following requirements: and are the internal and external temperatures of the room, and The minimum and maximum permissible room internal temperatures are h and c h is the equivalent thermal resistance and thermal capacity of the computer room; is the heat coefficient of the server, The energy efficiency coefficient of variable frequency air conditioning refrigeration; is the energy consumption of variable frequency air conditioner; is the maximum energy consumption of the inverter air conditioner; 3) Energy supply equipment constraints DCM's energy supply equipment includes RDG, CDG and energy storage. Except for RDG, all other equipment are controllable. Since CDG power generation is adjustable, CDG needs to provide a certain amount of backup capacity for the intraday stage in addition to power generation to cope with fluctuations in RDG and data center power loads: and is the minimum and maximum output of CDG, is the maximum climbing rate of CDG; Since the charging and discharging power of energy storage is adjustable, in addition to charging and discharging, energy storage can also reserve backup capacity for the intraday phase in the day-ahead phase, just like CDG. When considering energy storage to provide backup, the operating constraints of energy storage in the day-ahead phase are as follows: The maximum charging and discharging power of energy storage; and A binary variable representing the energy storage state. When it is 1, it means that the energy storage is in the charging and discharging state respectively. M is a sufficiently large constant. The state of charge constraint of energy storage can be expressed as: It is the energy storage charging and discharging efficiency; and The lower and upper limits of the energy storage charge state; When the power generation of CDG, RDG and energy storage in DCM is not equal to the load energy consumption, DCM can also ensure the day-ahead power balance by exchanging power with the upper main grid: and Provides day-ahead forecasting for WT and PV within DCM; 4) Carbon emission constraints DCM carbon emissions mainly come from purchasing electricity from the main grid and CDG power generation: W i CE is the total carbon emission of DCM i during the dispatch period, and are the carbon emissions from purchasing electricity from the main grid and generating electricity from CDG, where the carbon emissions from CDG are nonlinear functions. and is the secondary and primary carbon emission coefficient of CDG; b buy,E The primary carbon emission coefficient for purchasing electricity from the main grid; The carbon emission constraint of DCM in the day-ahead phase is expressed as: W i 0 is the initial carbon quota of DCM i; is the carbon quota exchanged between DCM i and DCM j. If it is greater than 0, it means that DCM i shares the carbon quota with DCM j. The subscripts i and j are the numbers of DCMs, which are used to distinguish different DCMs. CA is the carbon quota margin coefficient; W i buy and W i sell are the carbon allowances purchased and sold by DCM i, respectively; (2) Intraday constraints include load processing and equipment energy consumption constraints, energy supply equipment regulation and power balance constraints: 1) Load processing and equipment energy constraints The fluctuation of data load mainly comes from the prediction error of IN load, and the IN load processing adjustment amount The following constraints are met: Data load regulation will also cause fluctuations in energy consumption of various devices in the data center, and server energy consumption fluctuations The calculation method is as follows: and They are the energy consumption fluctuations of the data replication processing and receiving equipment, respectively; In order to ensure the stability of the room temperature, the amount of adjustment of the air conditioning cooling power during the day needs to offset the fluctuation of server heating: It is the adjustment amount of refrigeration and air conditioning power consumption during the day; 2) Energy supply equipment regulation and power balance constraints CDG and energy storage can adjust the output power within the reserve capacity range. The intraday power regulation constraints of CDG and energy storage are as follows: and is the uplink and downlink adjustment amount of CDG power; and It is the uplink and downlink regulation amount of charge / discharge power; In addition, power balance constraints must be met during the intraday phase: Step 3: Use piecewise linearization method to linearize the nonlinear constraints contained in the constraints in the scheduling model; The nonlinear constraints contained in the constraints of the scheduling model are mainly the carbon emission constraints of each DCM, which are linearized by piecewise linearization and expressed as: About The convex function of , so the above formula is further relaxed to: exist Take S+1 points on the function curve with equal horizontal coordinates, and mark the coordinates as (p s ,w s ), where s = 0, 1, 2, ..., S. Connect all adjacent points with straight lines, then The curve is approximated by S line segments; therefore, the carbon emission constraints of each DCM are further transformed into S linear inequality constraints, where the sth constraint is: Step 4: Use the binary variable expansion method to perform equivalent transformation on the ellipsoid uncertainty set. The detailed steps are as follows: Using uncertain variables represents the total RDG prediction error, which is the sum of WT and PV prediction errors, The uncertain set is expressed as: and As auxiliary variables representing the fluctuation range of RDG prediction error, since the RDG output fluctuation is the largest in the worst scenario, the above two variables can be processed as binary variables; and is the maximum upward and downward fluctuation range of the RDG prediction error, after considering the correlation between WT and PV prediction errors in step 2, and Use the CPLEX solver to solve the problem according to the following formula: Introducing binary variables and After that, the equivalent box uncertainty set of IN load can be expressed as: Step 5: Taking the minimum overall dispatch cost of all DCMs as the objective function, while satisfying the consistency constraints of the data load to be processed and the carbon quota interaction between DCMs, the ADMM algorithm is used for optimization and solution. On the basis of comprehensively considering the interaction between the data load to be processed and the carbon quota of DCMs, the asymmetric Nash negotiation model is used for profit distribution. The detailed steps are as follows: The objective function is to minimize the overall scheduling cost of all DCMs, while satisfying the constraints of the data load to be processed and the interaction consistency of carbon quotas between DCMs: x i is the variable for independent decision making of DCM i, v i,j is the coupling variable between DCM i and DCM j, specifically the data load to be processed for interaction between DCMs and interactive carbon quotas N DCM is the DCM set; In order to ensure the privacy of each subject, the ADMM method is used to solve the collaborative optimization scheduling problem of multiple DCMs. The coupling variables v are exchanged between each DCM. i,j The optimization result is used to update the local sub-problem model and continuously iterate the solution; when the coupling variables exchanged by adjacent DCMs are close enough, the iteration is completed and the best scheduling solution can be obtained. In the kth ADMM iteration, the optimization scheduling problem of the relaxed DCM i is expressed as: ρ is the quadratic penalty factor of the augmented Lagrangian function, is the average value of the coupled variable in the kth iteration, which can be updated as follows after k iterations: is the boundary variable v i,j The optimization result after the kth iteration, λ i,k is the Lagrange multiplier. To ensure the convergence of ADMM, λ i,k After each iteration, it is updated as follows: When the coupling constraints are satisfied within a certain error and the obtained results are stable enough, the original residual δ P and the dual residual δ D When both are less than the given value, the ADMM iteration process stops, δ P and δ D Calculate according to the following formula: The comprehensive contribution is calculated based on the data load to be processed and the amount of carbon quotas interacted by each DCM, and the profit is distributed using the asymmetric Nash negotiation model. The contribution of each DCM is r i The definition is as follows: γ DL and γ CA are the weights of the interaction between the data load to be processed and the carbon quota interaction, satisfying γ DL +γ CA =1; definition and are the operating costs of DCM i before and after it participates in the interaction between the data load to be processed and the carbon quota, For the benefits allocated to them through interaction, the profit distribution model based on the asymmetric Nash bargaining model is expressed as follows:

Images