Flexible operation method of VPP fusing IDC and BESS space-time interaction

Abstract

The application discloses a VPP flexible operation method fusing IDC and BESS space-time interaction, relates to the virtual power plant technical field, and solves the problem that the prior art lacks VPP optimal operation mode across time scales and spatial dimensions. The method comprises the following steps: in the day-ahead stage, according to day-ahead operation data of a power grid, a 24-hour benchmark scheduling plan is obtained by using a VPP day-ahead optimal scheduling model; the 24-hour benchmark scheduling plan comprises IDC task allocation, a BESS charging and discharging strategy and distributed power output; and in the day-in stage, according to the 24-hour benchmark scheduling plan and day-in operation data of the power grid, the charging and discharging power of the BESS and the real-time scheduling of IDC load are realized by using a VPP day-in dynamic rolling optimal model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual power plants, and in particular to a VPP flexible operation method fusing IDC and BESS time-space interaction. BACKGROUND

[0002] Under the background of low-carbonization and intelligentization of global energy structure, the power system urgently needs to cope with the volatility challenge of high proportion of renewable energy grid connection. As an innovative mode of aggregating distributed resources, virtual power plant (VPP) becomes a key technical path to improve the regulation ability of power grid by coordinating multiple flexible resources to participate in market transactions.

[0003] Internet data center (IDC) has unique time-space regulation potential: through the migration of computing power tasks and delay calculation, it realizes the load transfer in the geographical dimension, and uses the standby power system to complete the energy buffer in the time dimension, forming the dual regulation ability of "time elasticity" and "space elasticity", which is a high-quality demand response resource with high energy consumption elasticity, concentrated distribution and strong controllability.

[0004] Battery energy storage system (BESS) can provide flexible support for the system at the minute to hour level due to its fast response characteristics and bidirectional power regulation capability. When using renewable energy, BESS can first store the excess energy, and then convert these intermittent energy supplies into relatively uniform and stable power output during peak load periods. By participating in energy supply through BESS, the intermittency and instability of renewable energy generation can be solved, and IDC combined with BESS can realize the stable output of renewable energy.

[0005] However, existing researches mainly focus on the optimization scheduling of single type of resource, and lack of systematic exploration of the time-space coupling characteristics and collaborative mechanism of IDC and BESS. How to deeply explore the transferability and delayability of IDC computing tasks and the time-space complementarity of BESS charging and discharging strategies, and build a VPP optimization operation model across time scales (real-time regulation and day-ahead planning) and spatial dimensions (geographical distribution of resources) has become a key scientific problem to improve the economy and reliability of VPP, which has important theoretical value and practical significance for promoting the efficient configuration of flexible resources in new power systems and the deep integration of digital-energy systems. SUMMARY

[0006] In view of the above analysis, the embodiments of the present application aim to provide a VPP flexible operation method fusing IDC and BESS time-space interaction, to solve the problem of lack of VPP optimization operation mode across time scales and spatial dimensions in the prior art.

[0007] The present application discloses a VPP flexible operation method fusing IDC and BESS time-space interaction, which comprises:

[0008] During the day-ahead phase, based on the day-ahead operation data of the power grid, the 24-hour baseline scheduling plan is obtained by using the VPP day-ahead optimization scheduling model. The 24-hour baseline scheduling plan includes IDC task allocation, BESS charging and discharging strategy, and the output power of distributed generation.

[0009] During the intraday phase, based on the 24-hour baseline scheduling plan and the intraday operation data of the power grid, the VPP intraday dynamic rolling optimization model is used to realize the real-time scheduling of BESS charging and discharging power and IDC load.

[0010] Based on the above solution, the present invention also makes the following improvements:

[0011] Furthermore, in the VPP day-ahead optimization scheduling model, an objective function is constructed with the goal of minimizing the total VPP operating cost of IDC and BESS collaboration within 24 hours. Constraints are constructed that comprehensively consider power balance constraints, tie-line power constraints, reactive power balance constraints, and power computing coupling constraints. Based on the constraints, the objective function is solved to obtain the baseline scheduling plan for the 24 hours.

[0012] Furthermore, the total operating cost of a VPP in collaboration with IDC and BESS within 24 hours is calculated. The objective function with the minimum objective as its goal is expressed as:

[0013] (1)

[0014] in, Indicates the external power grid Electricity purchase cost during different time periods Indicating distributed power sources in Operating costs for a given period of time This indicates that BESS is in Operating costs for a given period of time Indicates that the system is in Cost of loss during a certain period.

[0015] further,

[0016] (2)

[0017] In the formula: Indicates in Electricity price during specific time periods In order to be in Power purchased from the external power grid during a given period; , , These are the coefficients for the quadratic term, the linear term, and the constant term of the operating cost of distributed power sources, respectively. Indicating distributed power sources in the output power of the time period; the unit cost of charging and discharging the energy storage, the charging power of the BESS in the charging power of the time period, the discharging power of the BESS in the discharging power of the time period; the cost coefficient of power loss, the loss power of the system in the loss power of the time period;

[0018] , , , , the decision variable of the VPP day-ahead optimization scheduling model.

[0019] Further, in the VPP intra-day dynamic rolling optimization model, a target function is constructed, which takes the weighted sum of the fluctuation of the distributed power source within 4 hours, the voltage deviation of the key node, and the risk value item as the target, and a constraint condition is constructed, which comprehensively considers the voltage sensitivity constraint and the key node screening condition constraint.

[0020] Further, in the VPP intra-day dynamic rolling optimization model, the rolling time window is T={t0, t0+Δt,..., t0+4h}, wherein Δt=15 minutes, a 16-time period optimization cycle is formed, and the target function is represented as:

[0021] (3)

[0022] In the formula, Pd(t) is the fluctuation of the distributed power source in the time period; Vd(t) is the voltage deviation of the key node in the time period; is the risk value item of the time period; is the weighting coefficient. Further,

[0023] is represented as:

[0024] (4)

[0025] wherein, Pd(t) is the output prediction of the distributed power source in the time period, is the response power of the demand side in the time period, which refers to the load adjustment power of the IDC; is the adjustment power of the BESS in the time period. ​​​​​​​

[0026] Further, is expressed as:

[0027] (5)

[0028] wherein, is a Monte Carlo expectation, is a 95% quantile risk value, is a conditional risk value; is a risk weighting coefficient;

[0029] In each optimization cycle, the corresponding Monte Carlo expectation is generated by a Monte Carlo method.

[0030] Further, is expressed as:

[0031] (6)

[0032] wherein, represents the total number of key nodes in the power distribution network; is a short-term predicted voltage value of node i in a time period; is a node voltage reference value; is a voltage adjustment amount of node i in a time period.

[0033] Further, the VPP day-ahead optimization scheduling model is solved by a solver, and the VPP day-ahead dynamic rolling optimization model is solved by an uncertainty robust optimization method.

[0034] Compared with the prior art, the present application can at least achieve one of the following beneficial effects:

[0035] The present application constructs a VPP flexible operation method integrating IDC and BESS, utilizes the time-space flexibility of IDC and the multi-scale regulation capacity of BESS, realizes the deep interaction of the two in time and space. By establishing IDC model and BESS model, a "time-space elastic coupling" mechanism is designed. A "day-ahead-real-time" double-layer optimization system is adopted, a market transaction mechanism and a reactive power sensitivity regulation strategy are integrated, multi-time scale resource matching is realized, day-ahead optimization takes cost minimization as the goal, and the real-time stage corrects fluctuations and reduces costs in 15-minute steps. At the same time, a robust decision-making method based on confidence interval is introduced to ensure stability and new energy consumption capacity under extreme scenarios, and the economic efficiency of VPP operation and the renewable energy consumption capacity are significantly improved.

[0036] The technical solutions in the present application can be combined with each other to realize more preferred combination solutions. Other features and advantages of the present application will be described in the following description, and some advantages will become apparent from the description, or will be understood by those skilled in the art through implementation of the present application. The purposes and other advantages of the present application can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0037] The accompanying drawings are included to provide a further understanding of the present application, and are incorporated in and constitute a part of this application, illustrate embodiments of the present application, and together with the description serve to explain the principles of the present application.

[0038] Figure 1 A flowchart of the flexible operation method of the IDC and BESS space-time interaction VPP provided by the embodiment of the present application;

[0039] Figure 2 A schematic diagram of the VPP and IDC, BESS collaborative optimization technical route framework provided by the embodiment of the present application;

[0040] Figure 3 A prediction information optimization scheduling diagram provided by the embodiment of the present application;

[0041] Figure 4 A schematic diagram of the working model of the IDC participating in the power grid energy management mode provided by the embodiment of the present application;

[0042] Figure 5 The load time transfer characteristics of the IDC provided by the embodiment of the present application;

[0043] Figure 6 A schematic diagram of power load regulation based on the space-time flexibility of the working load provided by the embodiment of the present application;

[0044] Figure 7 A circuit diagram of the ideal model of the BESS converter provided by the embodiment of the present application;

[0045] Figure 8 A schematic diagram of the typical vector four-quadrant operation state of the BESS provided by the embodiment of the present application;

[0046] Figure 9 A flowchart of robust optimization modeling considering uncertainty provided by the embodiment of the present application. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present application will be specifically described below in conjunction with the drawings, wherein the drawings constitute a part of this application, and are used to illustrate the principles of the embodiments of the present application, and are not used to limit the scope of the present application.

[0048] In one specific embodiment of the present application, a VPP flexible operation method integrating IDC and BESS space-time interaction is disclosed, and the flowchart of the method is shown in Figure 1. The specific implementation steps are described as follows:

[0049] In the day-ahead stage, according to the day-ahead operation data of the power grid, the 24-hour benchmark scheduling plan is obtained by solving the VPP day-ahead optimization scheduling model; the 24-hour benchmark scheduling plan includes IDC task allocation, BESS charging and discharging strategy, and output power of distributed power supply;

[0050] In the day-ahead stage, according to the 24-hour benchmark scheduling plan and the day-ahead operation data of the power grid, the charging and discharging power of BESS and the real-time scheduling of IDC load are realized by using the VPP day-ahead dynamic rolling optimization model.

[0051] Next, the implementation principle of the technical solution adopted in this embodiment is described as follows. Figure 2 The schematic diagram of the technical route framework of VPP and IDC, BESS collaborative optimization technology, covers the space coordination and time flexibility two-dimensional regulation and control mechanism. Figure 2 The technical route framework in can be divided into day-ahead layer and day-ahead rolling optimization layer. Among them, the day-ahead layer (time granularity is 1 hour) takes the minimization of 24-hour total cost as the target, optimizes the load baseline, unit combination and energy storage plan, and supports the day-ahead spot market transaction through the cross-regional price difference and standby diesel unit start-stop (response time 5-30 minutes). The core decision variables include calculating the load baseline, energy storage charging and discharging plan and refrigeration system load transfer strategy, which realizes economic scheduling by using the time sequence characteristics of new energy output and price difference. The day-ahead rolling optimization layer (time granularity is 15 minutes) takes distributed wind and light inverters, BESS and multiple IDC BESS as the regulation subject, and constructs a distribution network reactive power and voltage rolling optimization model. By screening the key nodes with high voltage sensitivity (sensitivity threshold > 0.05 p.u. / MW), the IDC energy storage fast response characteristics are preferentially used to correct the voltage deviation (voltage deviation is reduced by 1.2%) nearby, and based on the partition balance principle, the wind and light fluctuations are dynamically suppressed. This layer integrates multi-source prediction data (calculation load, photovoltaic output, price prediction) and Markov decision process to realize 15-minute step closed-loop rolling optimization, dynamically adjust the calculation load and energy storage charging and discharging, and improve the real-time cost optimization efficiency. At the same time, through the dynamic scene tree updating algorithm, the multi-time scale prediction information is fused, which significantly reduces the scheduling failure rate under the wind and light sudden change scene, realizes the new energy consumption rate improvement. The prediction information optimization scheduling diagram is as follows: Figure 3As shown, the upper-level scheduling corresponds to the day-ahead scheduling, and the lower-level scheduling corresponds to the intraday scheduling. The main improvements of the VPP and IDC, BESS collaborative optimization method provided in this embodiment are mainly reflected in: (1) The day-ahead layer integrates IDC computing power baseline (IDC power load), energy storage charging and discharging plan, etc., to realize the connection with the day-ahead spot market; the intraday layer dynamically adjusts the operation strategy through real-time data to quickly smooth out wind and solar fluctuations and correct deviations. It takes into account both market and real-time stability. (2) The combination of LSTM prediction and Monte Carlo scene generation (in the intraday rolling optimization, LSTM is first used to correct the prediction, and then Monte Carlo scene sampling is performed based on the correction value, and the risk quantification of multiple scenarios is integrated into the target and constraints. After each optimization is completed intraday, the system obtains new measurement data, LSTM re-predicts the error, Monte Carlo resampling is performed, a new scene tree is reconstructed and the next round of rolling optimization is started), and diverse uncertain scenarios are constructed online and dynamically optimized within the intraday rolling window to improve the robustness of the system. Compared with the static prediction model, the prediction deviation is greatly reduced. (3) Introduce reactive voltage zone dynamic control during the daytime phase, and combine model predictive control to correct voltage over-limit problems in real time.

[0052] Modeling the IDC. As the IDC becomes the main load in the VPP, its energy consumption characteristics are modeled in detail based on its power load characteristics. The IDC mainly consists of computing equipment and auxiliary equipment. Computing equipment accounts for a large proportion of energy consumption; while auxiliary equipment, used to ensure the reliable and stable operation of the data center, consumes less energy and can be considered a fixed value. The computing equipment in the IDC is the part of the data center used to process computing tasks, including several servers, and is the main hardware infrastructure of the IDC computing equipment. The power consumption model of the IDC is expressed as:

[0053] (1)

[0054] in, Indicates the first One IDC in Power consumption (kW) during the time period Indicates the first Power usage efficiency of an IDC Indicates the first Power consumption (kW) of auxiliary equipment in each IDC Indicates the first The computing devices of each IDC are The power consumption (kW) during a given period is directly related to the number of active servers in the IDC.

[0055] (2)

[0056] in, Indicates the first The growth coefficient of the computing power consumption of the servers in the IDC reflects the mapping relationship between the number of server activities and the computing power consumption. As the number of servers increases, the dynamic power consumption also increases accordingly. represents the computing power consumption of the active servers of the th IDC in the time period. represents the fixed base power consumption of the servers in the th IDC.

[0057] (3)

[0058] wherein, represents the number of active servers of the th IDC in the time period, represents the computing power consumption of a single server in the IDC.

[0059] For the working mechanism of the IDC and the aggregation mode thereof in the VPP, please refer to the working model of the IDC in the power grid energy management mode diagram in Figure 4 . The load time transfer characteristics of the IDC are shown in Figure 5 . By dynamically scheduling the processing time sequence of different types of computing tasks, the IDC can reshape the power consumption curve and realize the time shift optimization of the power load, so as to adapt to the peak-valley electricity price of the power grid and reduce the energy cost. The load scheduling management system dynamically adjusts the computing load distribution in the time and space dimensions, which significantly improves the electricity flexibility of the IDC. In the time dimension, the system can transfer the adjustable load in the peak period (such as delay-tolerant tasks) to the low power period for processing, so as to relieve the power supply pressure of the power grid; in the space dimension, the overload tasks are preferentially scheduled to the IDC nodes adjacent to the power grid, and the load transfer without physical line constraints is realized by using the optical communication network. Figure 6 The power load regulation diagram based on the time and space flexibility of the working load, Figure 6 The empirical results show that the peak load is successfully transferred across time periods among the three IDCs, in which IDC1 allocates 4 red loads (limited by capacity) to IDC3 and IDC2 according to the principle of power grid proximity, and only retains 1 yellow load associated with sensitive load for local processing. This time and space scheduling strategy makes the IDC become a new type of flexible resource, and therefore, the optimization model needs to simultaneously consider the power grid operation benefit and the IDC operation cost, and balance the technical feasibility and economy through multi-objective decision making, so as to finally build a power-information deep coupling collaborative optimization system. Formulas (4) to (16) establish the IDC “power-computation” time and space flexible transfer related models.

[0060] The working load change amount model of each IDC is represented as:

[0061] (4)

[0062] where, represents the net load change of the th IDC in the th period; represents the new workloads from the front-end types (such as Web, AI training, etc.) in the th period; is the business weight coefficient of the workloads of the front-end types ; represents the cross-region workloads migrated from the th IDC to the th IDC in the th period; is the migration success rate factor (dimensionless) subject to network latency and bandwidth; represents the terminal workloads that have completed computation in the th period; is the task completion efficiency coefficient, which is positively correlated with the performance of computing resources; represents the low-priority workloads suspended due to resource shortage in the th period; is the suspension decision threshold coefficient, which depends on the scheduling strategy.

[0063] The following is a cross-region migration decision model:

[0064] (5)

[0065] where, is the migration willingness coefficient of the th IDC (positively correlated with the load rate); is the workloads of the th IDC in the th period; is the network latency (ms) between the th IDC and the th IDC; is the network latency (ms) between the th IDC and the th IDC; is the delay sensitivity attenuation factor (ms⁻¹); is the real-time resource utilization rate (%) of the target th IDC; is the migration reception security threshold (usually set to 85%).

[0066] The work load fluctuation model of each IDC is represented as:

[0067] (6)

[0068] (7)

[0069] where: IDC total load amount in the time period; net load increment in the time period; online load instantaneous fluctuation amount (e.g. financial transaction request pulse) in the time period; periodic offline load increment in the time period; random non-periodic task increment in the time period; load type weight coefficient, dynamically adjusted by the scheduling strategy.

[0070] The energy efficiency coupling constraint is expressed as:

[0071] (8)

[0072] where: energy use efficiency (dimensionless) of the th IDC; rated load capacity of the th IDC; cumulative load total amount of the th IDC in the time period; , , respectively represent the constant term, the first-order term, and the quadratic term regression coefficient of the PUE curve, calibrated by measured data.

[0073] The work load start and end time period constraints are explained as follows, the system needs to satisfy the work load conservation law on the time domain boundary:

[0074] (9)

[0075] (10)

[0076] cumulative load total amount of the th IDC in the time period.

[0077] ​​​​​Equation (9) enforces the workload conservation at the beginning and end of the period, ensuring the system dynamic balance; Equation (10) limits the physical feasibility of the load by non-negativity constraint, avoiding negative values and defining the lower bound of the optimization feasible region. Both of them provide a basic constraint framework for the load scheduling model.

[0078] The load work capacity constraint is expressed as:

[0079] (11)

[0080] In the formula: is the maximum value of the cumulative workload of the th IDC.

[0081] The IDC service quality constraint is expressed as:

[0082] (12)

[0083] In the formula: is the weighted service quality constraint value of the th IDC within the time window under the service level agreement (SLA) constraint. Delay Threshold represents the maximum delay time unit allowed.

[0084] (13)

[0085] (14)

[0086] (15)

[0087] (16)

[0088] In the formula: is the transfer calculation load of the th IDC in the period; is the transfer load of the th IDC to the th IDC in the period. is the execution calculation load of the th IDC in the period; is the execution type calculation load in the period; is the transfer type calculation load in the period; Binary variable for execution state, 1 for execution, 0 for non-execution; Binary variable for transition state, 1 for transition, 0 for non-transition.

[0089] Modeling of BESS is modeled. Without considering the circuit process of battery BESS device, it is mathematically modeled from the aspects of state of charge (SOC) and charging and discharging power.

[0090] Charging process:

[0091] (17)

[0092] Discharging process:

[0093] (18)

[0094] Wherein, SOC (t) represents the state of charge of BESS at time t; represents the self-discharge rate of BESS; , respectively represent the charging power and discharging power of BESS at time t; , respectively represent the charging efficiency and discharging efficiency of BESS; represents the rated capacity of BESS; is a sampling interval, which can be appropriately taken as 1 hour. The battery cost function of BESS is represented as:

[0095] (19)

[0096] In the formula: is the capacity size of the configured BESS, which is a decision variable;

[0097] is the daily line loss cost of the distribution network after the BESS is configured, which can be obtained by solving according to the operation strategy of the BESS; , respectively represent the daily investment cost and daily operation and management cost of the BESS; the BESS to some extent represents the peak load shifting ability of the BESS. The calculation formulas of each component are as follows: (20)

[0098] (21)

[0099] (22)

[0100] ​​

[0101] Where: m a This is the cost coefficient for active power line loss; In order to be in Active power line losses in the distribution network during the time period; Indicates different BESS; and The first The rated power and rated capacity of each BESS; These are power and energy cost coefficients, respectively. The service life of BESS; This represents the operating and management cost coefficient for BESS. For BESS The charging and discharging power during the time period.

[0102] Considering that the battery cannot be in both charging and discharging states simultaneously within a unit time interval Δt, the battery's charging and discharging states must meet the following constraints:

[0103] (twenty three)

[0104] In the formula, For BESS The charging state during a time period, Xt∈{0,1}; For BESS The discharge state during the time period, Yt∈{0,1}.

[0105] The number of charge-discharge cycles of a battery within a scheduling cycle affects battery life; therefore, the following constraints are added:

[0106] (twenty four)

[0107] In the formula, , These are the limits on the number of times the battery can be charged and discharged.

[0108] During the system's operational optimization process, the energy state of the battery must satisfy the constraint that it remains equal at the beginning and end of the scheduling cycle:

[0109] (25)

[0110] In the formula, , These represent the initial and final states of charge of the battery, respectively.

[0111] The circuit diagram of the ideal BESS converter model is as follows: Figure 7 As shown, in Figure 7 In the diagram, E and EL represent the grid-side voltage and load electromotive force, respectively; UL, U, and U0 represent the load electromotive force.dc These are the grid-side inductor voltage drop, AC measured voltage, and DC-side bus voltage, respectively; I, I dc These represent the grid-side current and the DC-side current, respectively; L is the grid-side inductance; and RL is the load resistance. Figure 8 This is a schematic diagram of a typical vector four-quadrant operation state of a BESS, illustrating the active and reactive power regulation capabilities of the BESS under different operating states. Figure 8 It can be seen that when the grid-side voltage E is used as a reference, by changing the AC measurement vector U to control the grid-side current vector I, the phase difference between the grid-side voltage E and the current I changes, and the ESS exchanges active and reactive power with the grid, realizing the reactive power compensation function. This embodiment adopts more accurate SOC dynamic modeling, and processes charging and discharging power separately; it considers the impact of the number of charge and discharge cycles on battery life; and it introduces a BESS four-quadrant flexible operation model to dynamically adjust the voltage and achieve precise control of active and reactive power.

[0112] Model the diesel generator, which is the backup power source inside the IDC.

[0113] Torque balance formula:

[0114] (26)

[0115] In the formula: J is the moment of inertia; ω is the angular velocity (related to frequency); This is the engine's output torque; denoted as generator load torque; B is the coefficient of friction.

[0116] Voltage formula for synchronous generator model (dq coordinate system):

[0117] (27)

[0118] In the formula: , These represent the d-axis and q-axis voltages of the synchronous generator model, respectively. These represent the d-axis and q-axis currents of the synchronous generator model, respectively. For excitation flux linkage.

[0119] Electromagnetic torque formula:

[0120] (28)

[0121] In the formula: p is the pole logarithm.

[0122] For the speed control system model, the fuel supply is adjusted to maintain a constant speed.

[0123] (29)

[0124] where f is the actual frequency, is the reference frequency.

[0125] For the excitation system model, the excitation current The generator output voltage V out is controlled by adjusting the excitation current

[0126] (30)

[0127] where: is the excitation constant.

[0128] Next, IDC and BESS are involved in the modeling of VPP multi-time scale operation regulation.

[0129] (1) VPP day-ahead optimal scheduling model

[0130] The VPP day-ahead optimal scheduling model provided in this embodiment can integrate the flexibility of IDC and BESS, integrate IDC, BESS, diesel engine and other multi-source equipment, generate a benchmark scheduling plan with the minimum cost as the target, and realize resource coordination under new energy output constraints; In the rolling optimization regulation target, the new energy fluctuation, the key node voltage deviation and the risk value are comprehensively considered. In the VPP day-ahead optimal scheduling model provided in this embodiment, in addition to the conventional constraints, power-computing power coupling constraints are also added.

[0131] The objective function and constraint conditions of the VPP day-ahead optimal scheduling model are as follows.

[0132] The objective function is constructed with the lowest total operation cost of IDC and BESS coordinated VPP in 24 hours as the target. The total operation cost of IDC and BESS coordinated VPP is expressed as:

[0133] (31)

[0134] wherein, represents the power purchase cost of the external power grid in the time period, represents the operation cost of the distributed power supply in the time period, represents the operation cost of the BESS in the time period, represents the loss cost of the system in the time period. Among them, , are directly related to IDC.

[0135] (32)

[0136] wherein: denotes the electricity price at , is the power purchased from the external grid at ; , , are the quadratic, linear and constant coefficients of the distributed power generation cost, respectively, denotes the output power of the distributed power generation (including wind turbines and photovoltaics) at ; is the unit cost of energy storage charging and discharging, is the charging power (kW) of the BESS at ; is the discharging power (kW) of the BESS at ; is the cost coefficient of power loss (yuan / kWh), is the loss power (kW) of the system at , the power loss of the system mainly includes line power loss, device charging and discharging loss, etc. The real-time task adjustment of IDC (such as postponing tasks or migrating to other nodes) can quickly respond to grid fluctuations, optimize power distribution, and reduce additional losses caused by power imbalance.

[0137] In this embodiment, , , , , are the decision variables of the VPP day-ahead optimization scheduling model, and the IDC task allocation is performed according to and , and the BESS charging and discharging strategy is generated according to , , is the output power of the distributed power generation.

[0138] Next, the constraint conditions in this embodiment are specifically explained as follows.

[0139] 1) Power balance constraint

[0140] (33)

[0141] In the formula: is the power generation of the wind turbine at ; is the power generation of the photovoltaic at ; ; is the load other than IDC at The power consumed by the VPP, other loads such as electric vehicles and residential loads, are treated as a fixed value in the optimization model because the scheme focuses on IDC and BESS.

[0142] 2) Tie line power constraints

[0143] (34)

[0144] In the formula: For nodes in the distribution network exist Total power supplied by the power grid during the period For nodes in the distribution network exist Total load power during the period In order to be in Time period from node via transmission line to node The power; This represents the total number of nodes in the distribution network; .

[0145] Tie-line power constraints are essentially feasible region constraints; they don't appear directly in the objective function but indirectly affect its optimal value by "restricting the range of decision variables." This formula describes the power balance in a power grid. In any... Time period, node The total power supplied by the power grid minus the load power of a node should equal the sum of the power transmitted through all connected lines to that node. This reflects the power flow and balance in the power grid and is an important concept in power grid analysis and optimization.

[0146] 3) Reactive power balance constraint

[0147] The reactive power balance constraint is expressed as:

[0148] (35)

[0149] In the formula: For BESS Time period, at the node reactive power output, For nodes exist Total reactive power demand during the period For the line exist Reactive power transmitted during a specific time period.

[0150] By adjusting the reactive power output, BESS can correct the node. The voltage deviation should be controlled, and the voltage should be kept within the allowable range. Ensure the node... To achieve reactive power supply and demand balance and avoid voltage over-limit or power factor problems.

[0151] 4) External power grid purchase constraints

[0152] In addition, the amount of active or reactive power purchased from the external power grid must be less than the maximum value shown in formulas (36) and (37).

[0153] (36)

[0154] In the formula: for The maximum active power (kW) that the main grid is allowed to transmit to the VPP during a given time period.

[0155] (37)

[0156] In the formula: for The maximum reactive power (kVar) that the main grid is allowed to send to the VPP during a given time period.

[0157] 5) Branch power constraints

[0158] Branch power constraints are related to the calculation of active and reactive power flowing in the line, which are nonlinear formulas.

[0159] (38)

[0160] (39)

[0161] In the formula: , The lines are respectively The conductivity and susceptance , They are nodes exist Voltage amplitude and phase angle during the time period , The lines are respectively exist Active power and reactive power transmitted during a time period.

[0162] Formulas (40) and (41) specify the active and reactive power limits for the line.

[0163] (40)

[0164] (41)

[0165] In the formula: The alternative value of the line power limit (a very large positive number) is replaced by the line rating (kW / kVar) in practical applications.

[0166] 6) Node voltage constraint

[0167] Considering the large number of samples, the line constraint condition is removed, so there is no possibility of congestion and overload in the line. The constraints related to the minimum and maximum values of voltage deviation are shown in equation (42)

[0168] (42)

[0169] In the formula: is the node The voltage in the time period.

[0170] (7) Power-computing coupling constraint

[0171] (43)

[0172] In the formula: is the The basic computing load in the time period unrelated to refrigeration, ΔP cool,t is the The refrigeration system load transfer amount in the time period, which satisfies the transfer time window constraint.

[0173] Equation (43) works with other constraint conditions to ensure that the optimization model is physically and operationally feasible. Equation (43) considers two seemingly different scheduling lines, "computing space-time migration and power load", which defines the structure of the feasible solution and directly maps the trade-off of computing scheduling to the increase and decrease of power cost through the cost function, ensuring that the entire VPP scheduling problem can be efficiently and integrally solved by the mathematical model.

[0174] In the above VPP day-ahead optimization scheduling model solving process, the day-ahead operation data of the power grid includes wind power and photovoltaic output, day-ahead electricity price data, IDC initial computing load baseline, refrigeration load demand, BESS parameters, and node voltage reference value. In the specific implementation process, the VPP day-ahead optimization scheduling model can be solved by using a solver (such as a CPLEX solver), i.e., based on the above constraints, the 24-hour benchmark scheduling plan is obtained by solving the objective function based on the solver.

[0175] (2) VPP intra-day dynamic rolling optimization model with multiple types of resources

[0176] In the dynamic adjustment process of the VPP intra-day dynamic rolling optimization model, the decision variables are the charging and discharging power of the BESS and the IDC load scheduling; the distributed renewable energy output is collected, which is the basic data of the distributed power fluctuation; the load demand in the system is collected; and the system state parameters, i.e., the node voltage, are collected.

[0177] The rolling adjustment optimization objective function is described as follows.

[0178] Let the rolling time window be T = {t0, t0+Δt,..., t0+4h}, where Δt = 15 minutes, and a 16-period optimization cycle (window length is 4 hours, and each scheduling step is 15 minutes = 0.25 hours, 4 / 0.25 = 16) is formed, and the objective function is expressed as:

[0179] (44)

[0180] In the formula: is the fluctuation of the distributed power in the period; is the voltage deviation of the key node in the period (IDC, BESS node); is the risk value item in the period; is the weighting coefficient.

[0181] is expressed as:

[0182] (45)

[0183] wherein, is the output prediction of the distributed power in the period, is the response power of the demand side in the period, the demand side refers to the load side, the response ability of the flexible load side resource in the VPP to the power fluctuation, and the demand side response power in the scheme mainly refers to the load adjustment power of the IDC; is the adjustment power of the BESS in the period, which refers to the active power adjustment amount of the BESS, which is realized through charging and discharging, and is used for smoothing the renewable energy processing.

[0184] is expressed as:

[0185] (46)

[0186] In the formula: represents the total number of key nodes in the distribution network; is the voltage of node i in the The ultra-short-term predicted voltage value of the time period is calculated according to the day-ahead steady-state node voltage and the power / voltage sensitivity matrix, and the active / reactive power injection deviation of each node; is the node voltage reference value; is the node voltage reference value; is the voltage adjustment amount of the time period, which is a partition state variable.

[0187] is represented as:

[0188] (47)

[0189] In the formula: is the Monte Carlo expectation; is the 95% quantile risk value; is the conditional risk value; is the risk weighting coefficient, which is determined by prior sensitivity simulation.

[0190] At the beginning of each optimization period, the Monte Carlo expectation is recalculated according to the latest data. However, it is a constant in the optimization calculation of this period and does not change dynamically over time.

[0191] At every 15-minute time step, the risk value item is recalculated at each rolling period to dynamically reflect the latest uncertainty; the Monte Carlo expectation is generated by the Monte Carlo method, based on LSTM prediction and historical data, to calculate the expected value of the cost of each scenario. Specifically, the Monte Carlo method generates multiple possible scenarios for the operation of IDC and BESS within the next 4 hours, including power output, load demand and risk factors, based on LSTM prediction and historical data. Relationship with the research object: These scenarios serve as input data for the optimization problem, supporting the calculation of Monte Carlo expectation, VaR and CVaR, and indirectly guiding the optimization of decision variables. The 95% quantile of the current Monte Carlo sample loss is taken; All losses greater than the VaR sample are averaged. The use of LSTM prediction error compensation and Monte Carlo scenario generation calibrates data in intraday rolling optimization, indirectly affecting the accuracy of day-ahead prediction.

[0192] The relevant constraint conditions include: node voltage constraint, distributed wind power, photovoltaic reactive power output constraint, voltage sensitivity constraint and key node screening condition constraint. The node voltage constraint, distributed wind power, photovoltaic reactive power output constraint are existing constraints, which will not be described here.

[0193] 1) Voltage sensitivity constraint:

[0194] (48)

[0195] In the formula, is the reactive power sensitivity matrix, is the reactive power variation; is the active power sensitivity matrix, is the active power variation; , , represents the node voltage.

[0196] The dynamic calculation matrix and are introduced to update the sensitivity matrix in real time in combination with the intra-day rolling optimization (time step 15 minutes, rolling time window 4 hours).

[0197] 2) Key node screening condition constraints:

[0198] (49)

[0199] In the formula: is used to quantify the overall sensitivity of the reactive power of node i to the system voltage; is the threshold value of the voltage sensitivity.

[0200] As can be seen from the above, the intra-day operation data includes IDC computing power demand and task migration state, real-time SOC and reactive power output of BESS, node voltage, reactive power / active power sensitivity matrix, and loss power.

[0201] In the specific implementation process, a partitioned dynamic reactive power regulation strategy can be used to solve the intra-day model. BESS and distributed wind power and photovoltaic can quickly respond to reactive power compensation, and the node voltage of each feeder is adjusted back to a reasonable range through rolling adjustment. The specific process is as follows. It should be noted that this part of the process is for adjusting the voltage deviation of the intra-day model, and the process described is the complete and reasonable solving process.

[0202] 1) According to the actual data of the active / reactive power output of each node in the ultra-short term, the deviation from the day-ahead data is calculated to form the active / reactive power deviation vector of each node: ΔP=[ΔP1,ΔP2,…,ΔPN]T, ΔQ=[ΔQ1,ΔQ2,…,ΔQN]T.

[0203] 2) According to the reactive power / voltage sensitivity matrix, the voltage amplitude offset of each node is calculated, and the expected voltage value of each node is quickly calculated from the steady-state voltage distribution.

[0204] 3) According to the expected value or upper and lower limits of the voltage of each node, the voltage deviation value ΔV=[ΔV1,ΔV2,…,ΔVN]T is calculated, and the reactive power compensation demand Qcom=[ Qcom 1, Qcom 2, Qcom 3,…, QcomN]T is calculated according to the reactive power / voltage sensitivity.

[0205] 4) Optimize the dynamic reactive power compensation of BESS and distributed wind power and photovoltaic power in the partition to minimize the voltage amplitude deviation of each node after adjustment compared with the reference voltage.

[0206] 5) Considering the change of reactive power injection at the grid-connected point of BESS and distributed wind power and photovoltaic power, especially the difference in compensation depth or adjustment direction, it is easy to cause the voltage of other nodes to exceed the limit again. The above process can be repeated until the voltage deviation of each node is within the specified range.

[0207] 6) If the voltage deviation cannot be adjusted to a reasonable range, it means that there is a large deviation in the day-ahead steady-state regulation scheme, which needs to be revised. The same target as the day-ahead optimization is adopted, and the optimization period is the subsequent time domain without rolling adjustment. The steady-state regulation plan is reformed.

[0208] In the specific implementation process, the VPP daily dynamic rolling optimization model can be used to correct the day-ahead plan and optimize the voltage deviation and risk. The specific implementation process is as follows: according to the day-ahead optimization, a 24-hour baseline scheduling plan (IDC computing power baseline, BESS charging and discharging plan) is generated as the initial condition of daily optimization; the daily optimization collects real-time data (new energy output, IDC load, voltage state) to correct the deviation of the day-ahead plan. The day-ahead layer (1-hour granularity) provides long-term planning, covering a 24-hour cycle, focusing on optimizing cost and resource allocation. The daily layer (15-minute granularity) dynamically adjusts in a 4-hour rolling window to respond to wind and light fluctuations and load changes, and corrects the day-ahead plan. The daily optimization is closed-loop and rolling in 15-minute steps to dynamically adjust the computing power load and energy storage charging and discharging, improving the real-time cost optimization efficiency. The running results (voltage deviation, SOC change) of the daily optimization are fed back to the day-ahead model to update the baseline plan for the next cycle.

[0209] In the specific implementation process, the VPP daily dynamic rolling optimization model can be solved by using the uncertainty robust optimization method. The specific implementation process is as follows. A VPP multi-uncertainty robust optimization model is constructed to cope with the randomness difference of wind and light and IDC load, and to realize the robust optimization solution of VPP operation (solving the daily model). First, a joint probability box (p-box) model of wind and light output and IDC load is established to define the joint distribution confidence interval boundary. Then, based on the historical prediction error data, the confidence interval parameters are dynamically adjusted through the Bayesian update algorithm to realize the closed-loop calibration of prediction error. Finally, the p-box model is converted into a robust optimization constraint to ensure that the dispatching scheme meets the safety margin within the confidence interval, and by solving the robust optimization model, the adaptability of VPP to uncertainty is improved, and the system safety and economy are considered. The uncertainty parameter perturbation of the expected value is represented as follows:

[0210] (50)

[0211] (51)

[0212] where: is the average value of the uncertain risk value; is the maximum value of the uncertain risk value; is the minimum value of the uncertain risk value; is the variation range of the uncertain risk value.

[0213] In addition, the uncertainty variable can be described by the following formula. In this formula, is a random factor between 0 and 1.

[0214] (52)

[0215] The robust optimization model mainly considers the worst case, and in the VPP operation optimization related problem, the following uncertain variables are mainly focused on: active / reactive power of load, and renewable energy generation. Therefore, the VPP operation power balance constraint considering uncertainty is expressed as:

[0216] (53)

[0217] (54)

[0218] (55)

[0219] (56)

[0220] where: is the average output of wind power of node i at period t, is the average output of photovoltaic of node i at period t, is the discharging power of energy storage at node s at period t, is the charging power of energy storage at node s at period t, is the parameter reference value (such as the average prediction value of load / renewable energy), is the parameter fluctuation range (the maximum deviation of the actual value relative to the average value), is the actual parameter value, is the conservativeness coefficient (between 0 and 1), is the active power demand of IDC at period t (which can be dynamically adjusted), is the average power of conventional load of node i at period t (including IDC reference load), is the reactive power demand of conventional load of node i at period t (including IDC reactive power demand), Power supply to VPP from the main grid (for balancing power shortage).

[0221] Next, the robust optimization confidence interval coupling constraint mechanism is described. Robust optimization models parameter fluctuations through uncertainty sets (box, ellipsoid, polyhedron, etc.), and its conservatism is determined by the set geometry. The probability box (p-box) model in the box set is used to balance the model complexity and conservatism, and the robust constraint is constructed by defining the parameter fluctuation boundary, taking into account the computational efficiency and the robustness of the optimization scheme. Confidence interval coupling constraints are established for the p-box model:

[0222] (57)

[0223] Where PP is the joint probability box, is the confidence threshold (default 5%).

[0224] A collaborative robustness index is established to define the coordination degree of IDC task migration and BESS charging and discharging , which measures the flexibility of the system. A Markov decision process (MDP) is defined, where: the state space S=(wind power prediction error, IDC load deviation, BESS SOC); the action space A=(IDC task migration amount, BESS charging and discharging power); the reward function R(s,a)=-dispatching cost voltage out-of-limit penalty; confidence interval adaptive adjustment, online optimization of confidence interval parameters based on Q-learning algorithm , to achieve dynamic risk control.

[0225] The solution process of robust optimization is described as follows.

[0226] (1) Offline modeling and initialization. Input renewable energy output historical data, IDC load curve, BESS parameters and power market rules. Construct a joint probability box model through kernel density estimation (KDE) and initialize the confidence interval parameters.

[0227] (2) Online rolling optimization. Real-time data acquisition, obtain wind power / photovoltaic ultra-short-term prediction value, IDC real-time computing power demand and temperature control state. Multi-scenario generation, generate N sets of source-load fluctuation scenarios (including ±20% extreme fluctuation) based on Monte Carlo sampling. Joint robust domain solution, call mixed integer second order cone programming (MISOCP) solver to generate a dispatching strategy that takes into account economic efficiency and safety. Markov decision optimization: select the optimal action atat according to the current state stst, update the confidence interval parameters.

[0228] (3) Execution and feedback. IDC task migration instructions and BESS charging and discharging plans are issued, and synchronous participation in power spot market and frequency modulation market bidding is performed. The system operation state (voltage limit rate, SOC change, etc.) is monitored and fed back to the dynamic calibration module to update the model parameters.

[0229] Figure 9 A flow chart of robust optimization modeling considering uncertainty is given.

[0230] In summary, the embodiment proposes a virtual power plant (VPP) flexible operation strategy that integrates the space-time interaction of data centers (IDC) and battery energy storage systems (BESS). To overcome the shortcomings of existing research on the coordination mechanism of IDC computing power migration and BESS multi-scale regulation capacity, a "space-time elastic coupling" collaborative scheduling model is constructed. First, a data center power consumption model and a space-time migration ability model are constructed to quantify its flexibility. The geographical dimension load transfer and time domain buffer capacity of standby power are considered, and the dynamic SOC model of BESS is used to represent its cross-period energy transfer characteristics, and a collaborative optimization framework is established. Second, a "day-ahead-real-time" two-level optimization system is designed, which integrates market trading mechanism and reactive power sensitivity regulation strategy, realizes multi-time scale resource matching, and reduces the intraday scheduling cost. Further, a robust decision-making method based on confidence interval coupling is proposed, which dynamically calibrates the IDC-BESS joint regulation boundary to reduce the system voltage limit rate and wind and light curtailment rate under extreme scenarios.

[0231] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. Among them, the computer readable storage medium is a disk, an optical disk, a read-only memory or a random access memory, etc.

[0232] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A VPP flexible operation method for fusing IDC and BESS space-time interaction, characterized in that, The method comprises: In the day-ahead stage, according to day-ahead operation data of the power grid, a 24-hour benchmark scheduling plan is obtained by solving a VPP day-ahead optimization scheduling model; the 24-hour benchmark scheduling plan comprises IDC task allocation, BESS charging and discharging strategy and output power of the distributed power supply; In the day-ahead stage, according to the 24-hour benchmark scheduling plan and day-ahead operation data of the power grid, the charging and discharging power of the BESS and the real-time scheduling of the IDC load are realized by using a VPP day-ahead dynamic rolling optimization model; In the VPP day-ahead optimization scheduling model, a target function is constructed, which aims to minimize the total operation cost of the VPP in 24 hours in cooperation with the IDC and the BESS, and constraint conditions are constructed, which comprehensively consider power balance constraints, tie-line power constraints, reactive power balance constraints and power and computing power coupling constraints; the 24-hour benchmark scheduling plan is obtained by solving the target function based on the constraint conditions; Total VPP operating cost over 24 hours with IDC and BESS coordination The objective function with the lowest target is expressed as: (1) wherein, represents the purchase cost of electricity from the external grid during the time period, represents the operation cost of the distributed power supply during the time period, represents the operation cost of the BESS during the time period, represents the loss cost of the system during the time period; In the VPP day-ahead dynamic rolling optimization model, a target function is constructed, which aims to minimize the weighted sum of the fluctuation of the distributed power supply in 4 hours, the voltage deviation of the key node and the risk value item, and constraint conditions are constructed, which comprehensively consider voltage sensitivity constraints and key node screening condition constraints; In the VPP day-ahead dynamic rolling optimization model, the rolling time window is T={t0,t0+Δt,...,t0+4h}, wherein Δt=15 minutes, a 16-period optimization cycle is formed, and the target function is represented as: (2) In the formula: In order to be in Fluctuations in distributed power generation over a given period; For key nodes Voltage deviation over a period of time; for Value at Risk (VaR) for a given period; These are the weighting coefficients.

2. The VPP flexible operation method fusing IDC and BESS space-time interaction according to claim 1, characterized in that, (3) In the formula: represents the electricity price in , is the power purchased from the external power grid in ; , , are the quadratic term coefficient, the linear term coefficient and the constant term coefficient of the distributed power operation cost respectively, represents the output power of the distributed power in ; is the unit cost of energy storage charging and discharging, is the charging power of the BESS in , is the discharging power of the BESS in ; is the cost coefficient of power loss, is the loss power of the system in . , , , , are decision variables of the VPP day-ahead optimization scheduling model.

3. The VPP flexible operation method of claim 2, wherein, is represented as: (4) wherein, is the output prediction of the distributed power source in is the output prediction of the distributed power source in is the response power of the demand side in is the response power of the demand side in is the adjustment power of the BESS in is the adjustment power of the BESS in 4. The VPP flexible operation method of claim 3, wherein, is represented as: (5) wherein, is the Monte Carlo expectation, is the 95th percentile risk value, is the conditional risk value; is the risk weighting factor; In each optimization cycle, the corresponding Monte Carlo expectation is generated by using the Monte Carlo method.

5. The VPP flexible operation method of claim 4, wherein, is represented by: (6) wherein, denotes the total number of critical nodes in the power distribution network; is the voltage reference value for node i; is the ultra-short term predicted voltage value for node i at time period t; is the voltage reference value for node i; is the voltage adjustment value for node i at time period t; is the voltage adjustment value for node i at time period t.

6. The VPP flexible operation method of claim 5, wherein, The VPP day-ahead optimization scheduling model is solved by using a solver, and the VPP day-ahead dynamic rolling optimization model is solved by using an uncertainty robust optimization method.

Images